Cannabis Sativa

thcscience_admin
Page 1
Social Robot Modelling of Human Affective State
D.B. Skillicorn, N. Alsadhan, R. Billingsley and M.-A.- Williams
May 3, 2017
Abstract
Social robots need to understand the affective state of the humans with whom they interact. Successful
interactions require understanding mood and emotion in the short term, and personality and attitudes
over longer periods. Social robots should also be able to infer the desires, wishes, and preferences of
humans without being explicitly told.
We investigate how effectively affective state can be inferred from corpora in which documents are
plausible surrogates for what a robot might hear. For mood, emotions, wishes, desires, and attitudes
we show highly ranked documents; for personality dimensions, estimates of ground truth are available
and we report performance accuracy. The results are surprisingly strong given the limited information
in short documents.
1 Introduction
Social robots must build and use models of the humans with whom they interact. The quality of these
interactions is much improved if robots are able to estimate the affective state of such humans.
Affective state can be considered as being made up of the following components, about which there is
wide agreement among psychologists [4]:
• Mood. This captures a basic state of mind that is internally generated (although the drivers are
not well understood) and relatively long lasting. It can be divided into two components: positivity
and negativity, which are independent rather than opposites. In other words, an individual can have
increased positivity, without necessarily having reduced negativity. Mood acts to dampen the intensity
of countervailing emotions.
• Emotions. Emotions are a parallel channel to cognition that focuses on the values associated with
external situations. Studies of humans with brain injuries that reduce emotional intensity indicate
that such injuries cause difficulty making decisions, suggesting that a critical part of the function of
emotions is to rank the importance or significance associated with potential goals and tasks. Emotions
are primarily a reaction to external circumstances. There are a number of categorization of basic
emotions, but we will use anger, disgust, fear, and sadness (which are considered negatively associated
emotions); and anticipation, joy, and surprise (which are considered positively associated emotions).
High positive mood tends to enhance the expression of positive emotions and dampen the expression
of negative emotions, and the converse.
• Personality. One definition of personality is “stable differences in motivational reactions to circum-
scribed classes of environmental stimuli” [5]. In other words, personality captures the toolkit of re-
actions (often called “reaction norms”) that each individual preferentially uses. Personality is often
quantified in terms of the so-called Big Five dimensions, with acronym OCEAN: openness, conscien-
tiousness, extraversion, agreeableness, and neuroticism. Each dimension actually represents an opposite
pair, for example extroversion-introversion, and an individual is placed on axes corresponding to their
expression of each of these five property pairs. It is typically assumed that personality is fixed (at least
by adulthood) for each individual.
1
arXiv:1705.00786v1 [cs.RO] 2 May 2017
• Values. These associate positivity or negativity with both physical and abstract objects: concepts,
beliefs, and goals. They are primarily cognitive.
• Attitudes. These are evaluations of specific instances of people, ideas, events, situations, and other
more-concrete objects, providing shorthand guidelines for the proper responses to them, including
potential actions. In an ideal world, attitudes would align with, and so be predictable from, values but
few individuals exhibit this level of consistency.
Figure 1: Relationship among components of affective state
The relationships between these components of affective state are shown in Figure 1. Components on
the left are relatively fixed; they influence components further to the right, which are more volatile: values
and personality fixed over a major fraction of a lifetime, attitudes fixed over years, mood fixed over a day,
and emotions varying on a time scale of seconds.
Why should a robot take into account the affective state of a human with whom it is interacting? Social
interactions assume an understanding of affect by all parties, and some autism-spectrum disorder individuals
have difficulty in social settings precisely because they lack such understanding. A robot cannot be called
‘social’ without at least some awareness of the role of affect. The feedback loop that seems to be required
to create effective interactions, between human and human and now between human and robot, has been
called the affective loop [11].
In particular, the ability to infer a human’s emotions is useful for the following reasons:
• The robot can compute more accurately the priorities that the human will have for his/her own
behavior.
• The robot can improve its own prediction of the desires of the human, and so can make itself more
useful.
• The robot can detect when it is performing poorly from the human’s perspective.
• The robot can mimic the human more effectively to create better rapport.
• The robot can act to help alter the human’s emotional state, when appropriate.
Because emotions change rapidly, a robot’s model of a particular human’s emotion must update frequently.
The process for inferring emotion must therefore be lightweight. Since emotions are volatile, they can only
be inferred from recent data about each human, and consequently from small amounts of data.
The reason to infer a human’s mood is that inference of emotions cannot work well without modelling
the effect that mood has on them. We might casually say that ‘X is an angry person’ but it would be more
accurate to say that ‘X’s is habitually in a strongly negative mood, and this amplifies his emotion of anger’.
It might be imagined that another motivation for a robot to infer a human’s mood might be to try
and alter it. When someone is in a ‘bad mood’ we often talk about trying to cheer them up. However,
technically, moods are only very weakly influenced by others or the environment. What we mean when we
talk of cheering someone up is really to act to reduce their emotion of sadness. The process of altering
another’s emotional state deliberately is quite a difficult one, even for humans; it may eventually be possible
to teach robots to do this, but it is beyond our present capabilities.
The reason to infer the human’s personality is that it delimits the repertoire of normal reactions that each
individual human will choose from in any particular situation, and therefore allows the robot to anticipate
2
more accurately. Different humans behave differently, even when they are in, say, the same bad mood and
feeling angry; and a social robot must be able to interact with each human as an individual.
Because personality is relatively fixed, a robot can infer a particular human’s personality based on the
totality of data about that human, perhaps refining it incrementally. More complex models can be built,
since they can be developed and improved as a background process.
Computational approaches to values and attitudes are not well developed except in contexts such as
marketing and advertising, where they are often further narrowed to attitudes towards particular products.
The difficulty with modelling attitudes is that they are really relations between an individual and external
objects, rather than an internal property of the individual. The ability to infer values and attitudes will be
useful to a robot in determining a human’s responses to the current context, and immediate intentions. We
explore the use of the semantic differential as a first step towards computational modelling of attitudes.
2 Modelling Strategy
We must use data that resembles what a robot might (over)hear in the presence of a human whom it is trying
to model, and analysis that can plausibly be carried out in an environment with limited time and resources.
For example, full parsing, or even part-of-speech tagging, may be too expensive, and perhaps not effective
given the fragmentary, non-grammatical nature of informal speech. For those properties for which models
must be built quickly, from small amounts of data, accuracy cannot be expected to be high. A human might
be able to infer the emotional state and mood of another human from a few utterances, but would typically
need more to infer personality.
Approaching affective state lexically has been a popular approach for almost a century. As a result, a
number of lexicons have been developed. In each case, word frequencies are considered to be signals of a
component of affective state. Empirical studies have shown that even emotional state – for which more than
500 distinguishable emotions have been suggested – can be described using a much smaller set of core or
underlying emotions, and this is characteristic of other aspects of affective state.
Components can be unipolar or bipolar. For example, the emotion of anger is unipolar: one can be
angry, but the opposite is a neutral state of not being angry. In this case, a count of words from a lexicon
of anger-related words estimates how angry the speaker or writer of an utterance is. Emotions are usually
regarded as unipolar from a psychological perspective (although in common parlance, many emotions have
named opposites, e.g. angry-calm).
The Big Five personality traits are usually conceived of as bipolar; openness is really a pair: openness to
experience versus closedness to experience. A lexicon for bipolar components differentiates words based on
which ‘direction’ they are associated with.
Our basic approach is to use lexicons associated with each property of affective state we are interested
in. These are:
• For mood, positive and negative lexicons from the NRC Emotion Lexicon [15].
• For emotions, the seven lexicons for anger, disgust, fear, sadness, anticipation, joy, and surprise from
the NRC emotion lexicon [15]. Plutchik [19] listed these as primary emotions (including trust, which
seems to be an attitude rather than an emotion, that is a relation rather than a state). There are many
overlapping sets of ‘basic’ emotions, many of them deriving from the circumplex model of Russell [21].
For a more physiological discussion of basic emotions, see Ekman [6].
• A lexicon derived from the Wishes dataset to see whether it is plausible for a robot to detect an
expression of a wish, that is an implicit command [9].
• A lexicon, developed by the authors, for preferences to see whether it is plausible for a robot to detect
desires/preferences in humans.
• A lexicon based on the Osgood scores of the most common 1000 words in English, capturing the
dimensions of good-bad, active-passive, and strong-weak [17]. These have been shown to underlie
attitudes across many cultures.
3
Table 1: Dataset properties
Number of Av. doc. length
Dataset
documents
(& with rare words
removed)
Essays
2469
577 (563.5)
Facebook posts
9918
10.9 (8.6)
Tweets
100,000
8.99 (8.99)
Headlines
1250
6.5 (< 1))
Video blogs
404
587 (480)
The relationship between personality and word usage has mostly been explored from the perspective of
how personality influences words [23, 1] and the related efforts to understand the underlying dimensions of
variation in word usage (that is, to justify the Big Five – sometimes Big Six – dimensions). Attempts to
predict personality from word usage (for example, [8, 7]) have used varying approaches to what is being
predicted (one dimension at a time or an entire personality) and what attributes are used, often adding
syntactic attributes such as n-grams to lexical attributes.
Others have studied the converse problem, methods for robots to express emotion in ways that seem
natural to humans (thus closing the other side of the affective loop) [13, 12].
Attitudes have received less attention, except for the vast literature on sentiment analysis – the attitudes
of individuals to particular products. More general approaches include Liu and Maes attempt to build a
model of the attitudes of other humans that could be used as an advice surrogate [14] and more linguistic
work on the mechanisms by which attitude differences are expressed in language [20].
3 Experiments
We use the following datasets in our experiments:
• A set of essays written by university students about their daily lives and thoughts, collected by Pen-
nebaker between 1997 and 2004 and labelled with personality classes [18].
• A set of Facebook posts collected by Stillwell and Kosinski using a Facebook application that imple-
mented the Big Five test using Costa and McCrae’s NEO-PI-R domains, along with other psychological
tests [3].
• A set of transcripts of video blogs (vlogs) from Youtube, labelled with personality impressions by
crowdsourcing [2].
• The first 100,000 tweets in the Microsoft Research Conversation Corpus, a set of 4.46 million tweets
from about 1.3 million conversations.
• A set of 1250 headlines from the BBC, labelled with both mood and emotions by human raters, and
used as a problem dataset for the Semantic Evaluation workshop in 2007 [25].
The sizes and properties of these datasets are shown in Table 1.
Facebook posts and tweets (and perhaps headlines) are good representations of what a robot might
overhear, and can plausibly be used to infer mood and emotions. The essays and Youtube datasets are
less useful because these documents were produced over longer time frames, during which emotions might
change.
Properties such as mood and personality are more stable over time, so the essays and Youtube datasets
are more useful for estimating them.
For each of these datasets and each lexicon, we build a document-word matrix describing the counts of
the words in each document.
4
Variations in document length are an issue for all corpora, because of their effect of document-document
similarity. In the essays corpus, documents are of similar lengths; the Facebook posts vary in length from a
single word to several hundred words; tweets have an imposed upper bound of 140 characters, but we observe
a non-trivial number of very short tweets; headlines are all very short; and the Youtube vlogs are longer but
have significant length variation.
When normalizing to take document lengths into account, we address two underlying processes. One is
the occurrence of repeated words, which are far more common in tweets and Facebook posts than they would
be in ordinary speech, apparently as a mechanism for emphasis. To compensate, we take the logarithms of
word counts. This, for instance, reduces the count of positive or joyful words in a document like ”I’m happy,
happy, happy” in comparison to ”I’m happy, I’m ecstatic”, in which the expression of emotion has required
more cognitive processing.
The second is that the average word counts of longer documents are inevitably longer than the word
counts of shorter documents. The conventional way to address this is to divide each word count by the
length of the document in which it appears. This works reasonably well for documents of modest length
but, for short documents, the denominator is small and the resulting word rates become distortingly large.
To compensate, we divide word counts in a document by the digamma function of the total number of
words it contains. The digamma function has the shape of a logarithm but rapidly becomes much flatter.
It quite accurately fits the shape of the graph of new words encountered as a function of words encountered
in ordinary documents [24]. Its Taylor series expansion to a few terms is also quite a good approximation,
so it can be computed inexpensively. Applying this transformation modestly increases the weights of words
that appear in short documents while having little effect on words that appear in longer documents.
The reader is cautioned that Tweets and Facebook posts are presented intact, and so contain inappropriate
language.
3.1 Mood
Posts are processed to extract word counts for words associated with positivity and negativity lexicons,
these word counts are normalized for document length and scores are computed by summing the normalized
word weights for each document. This is, surprisingly, more effective than conventional techniques such as
computing the singular value decomposition of the document-word matrices.
The only corpus of which we are aware that is labelled with multiple emotions is the headlines corpus. It
was labelled by humans, but with only weak agreement between human raters [25]. Attempts to predict the
human-generated labels fail because they clearly made their judgements without using many of the words
of the NRC emotion lexicon. The problem seems to be not that the NRC lexicon is deficient, but that the
human raters made their judgements using considerable contextual knowledge rather than the words that
actually appear in the headlines. We provide comparisons of the top-ranked headlines by lexicon score and
by human rater for each emotion. Both approaches do not fare well (possibly because headlines are highly
stylized text).
For unlabelled data, we show ranked lists of the most highly ranked documents with respect to the
property being considered. Table 2 shows top-scoring positive mood tweets while Table 3 shows top-scoring
positive mood Facebook posts. Scores for tweets are integers because they are derived from word counts;
scores for Facebook posts are non-integers because word counts have been changed to word pseudo-rates by
document length normalization. Table 4 shows the top-scoring positive mood headlines both as ranked by
human raters, and using the word counts from the mood lexicon. The human scores are given as an average
value between +100 and −100, from most positive to most negative.
Table 5 shows top-scoring negative mood tweets, while Table 6 shows top-scoring negative mood Facebook
posts. Table 7 shows the most negatively ranked headlines, again based on both human raters and the lexicon.
For the essays dataset there is no need to normalize for document length, since all of the essays are
relatively long, and of similar lengths. Empirically, normalizing for repeated words seems to improve the
quality of the results slightly. Since positivity and negativity are not anti-correlated, the same essay can
score highly for both positivity and negativity. Essays are too long to include in their entirety, but some
example extracts are shown in Table 8.
5
Table 2: Top positive mood tweets. The limitations of a purely lexical approach can be seen in the 3rd
ranked tweet, where words like ’truth’ and ’home’ increase the positivity score when, from the context, they
should not. Note that strategies like calculating the ’net positivity’ are not valid because positivity and
negativity are independent properties.
Score Tweet
7
kinda... probably got a job... a good job too... a legit honest to god good job... you should ask
me what it is :] #fb
7
RE: High Road - Glad to help. U pray I pray We all pray. God provides. Our God is an awesome
God! keep on keepin on
7
user11571 truth hurts buddy. We deserved it in 95 with 20 home and away wins. Good luck with
the dance action.
7
rock a by baby on the tree top, when the wind blows the craddle will rock lol! or maybe think
good thoughts, like ya beautiful!
6
want to eat some rich chocolate icecream covered with decadent chocolate sauce with whipped
cream and a cherry on top and rainbow sprinkles!
6
- morning and happy national day! good nite sleep often lead to a good start for the day. enjoy
the rest of your day!
6
It won’t be long until you find love, and by love, I mean TRUE love. You are such a beautiful,
sweet, intelligent,
6
VERY SWEET of you! Thank you for the compliment. Rock on with Peace, Love, & Respect
Always!
The video blog transcripts have similar properties. The documents are longer, and with similar lengths,
so we normalize only to discount repetitions. The highest scoring positive and negative mood documents
are shown in Table 9.
3.2 Emotions
Documents are processed in exactly the same way using lexicons for three positive emotions: anticipation,
joy, and surprise; and four negative emotions: anger, disgust, fear, and sadness. Tables 10–29 show the top
scoring tweets, Facebook posts, and headlines for each emotion (except that anticipation was not included
in the Semeval data). We appeal to face validation, demonstrating that the highly ranked examples for each
dataset and each emotion do, on the whole, appear to exhibit that emotion strongly.
3.3 Wishes
Goldberg et al. investigated whether wishes or desires could be detected lexically. Their goal was to go beyond
the limitations of sentiment analysis (how does a consumer feel now?) to find objects that a consumer might
feel positively towards even if they were not actually present. They developed models for the language of
wishes based both on words themselves, and a set of templates (e.g. ‘I just want ...’).
We created a wish lexicon from their corpus in which each document is labelled as embodying a wish
or not. We extracted all parts of speech except nouns, and selected words based on their significance as
predictive attributes for the wish/non-wish class label. This produced a set of 28 words that are associated
with the wish/non-wish distinction. The lexicon is shown in Table 30.
We then repeated the scoring process used for measuring emotions with this new lexicon, getting word
counts for each document, normalizing for document length as before, and summing the adjusted counts.
Table 31 shows the top-ranked tweets and Table 32 the top-ranked Facebook posts.
3.4 Preferences
The same process was repeated with a lexicon manually constructed to capture aspects of desire and pref-
erence. The lexicon of desires and preferences is shown in Table 33. The difference between wishes and
6
Table 3: Top positive mood Facebook posts. The 7th ranked post is quite negative, but also contains strong
positive elements.
Score
Post
2.0019 Let me know you more deeply and truely, oh Lord. There are none who know you completely,
and so I pray that YOU show me. Your ways are perfect and I want my life to sing your praise.
Let me walk with love, joy, peace, patience, kindness, goodness, gentleness, and self-control.
Against such things there is no law.
1.8461 Hallelujah! Hallelujah! Holy! Holy! Are you Lord God almighty! Worthy is the Lamb, worthy
is the Lamb that was slain! Unto you be all glory and honor and praise! May my life bring glory
to your name :)
1.6792 Define me urban dictionary The Most wonderful person in the World. Kind, Sweet, Loving,
Caring, Gentle. Perfect in Every way. The one you love for all your life. Crazy hot girl.
Beautiful, smart & funny; *PROPNAME* posesses atributes absent in 99.9% of women. Truly
a lucky find. Plus she rocks. The most wonderful drug in the world, better know as lortab.
taking the pill may cause a sense of euphoria, & well being. hahaha
1.4489 The star burnned like a flame, pointing the way to God, the King of kings; the wise men saw
the sign and brought their gifts in homage to their great King. Seeing the star, the wise men
said: This must signify the birth of some great king. Let us search for him and lay treasures at
his feet: gold, frankincense, and myrrh.
1.3207 is wishing a Happy Birthday to my bestie *PROPNAME*! I love you honey, and I hope you
have a wonderful 22nd Birthday!
1.3181 The Soup: The Verdict Type: Vegetable Appearance: Fantastic Texture: Good Taste: Poor
Overall: Edible but not pleasant Summary: This is a great soup for looking at, it is full of
colour, healthy vegetables and creaminess! If you are not hungry make this soup and you will
feel good! However, if eating is your thing, this soup should be avoided. I have soup left for 3
meals - any takers? Only p&p to pay! Enjoy!
1.3024 Please keep *PROPNAME*’s family in your prayers tonight and tomorrow. We have to be in
court at 9am. Really pray for the oldest *PROPNAME* went to his dad’s for his summer visit
and had his mother, brother and sister stripped from him for six months. CPS is providing
Christmas for the other two but not him. I sure hope we succeed at stripping them of their
immunity and make them pay for what they’ve done to this family.
1.2794 Two weddings, one baby shower and a great church service on Sunday! A weekend full of love
and encouragement and I know the best is yet to come! All things work together for the good
to them that love God, for them who are the called according to his purpose.
1.2473 money, teeth cleaning, g-ma’s, bus, top secret errand, target, rodarte dress for $40(!!!), some
warm and comfy- themed xmas shopping, spending time with mom, coffee, sephora.com, the
office, stepping outside and forgetting how cold it gets without proper coverage- productive and
selfindulgent bliss. not a bad day off.
1.2473 Feeling DAMN GOOD headwise lately: forceful self-change of outlook, meds seem to be finally
working, & I’m startin to get shit done! ’Course, meds need tweaking–one of ’em (dunno which)
is also tryin to make me a zombie–not the fun movie kind, either: I get sleepy around 4 or 5 PM
now; too damn early even if I weren’t a night owl! PSYCHED bout the forward progress, but
this zombie thing’s crampin’ my nightlife, oy!
preferences is that preferences usually refer to either the current, or imminent, situation, while wishes usu-
ally indicate a longer-range desire or preference. A robot might process a desire or preference immediately,
while storing a wish as part of its model of the corresponding human.
Table 34 shows the top-ranked tweets and Table 35 the top-ranked Facebook posts for this model.
7
Table 4: Top positive mood headlines. The human-selected headlines require considerable context to under-
stand why they are positive (and the first one remains mysterious); the lexically ranked headlines struggle
with the extremely short documents.
Method
Score Post
87
Goal delight for Sheva
Human
85
Pamuk wins Nobel Literature prize
84
Amazon.com has ’best ever’ sales
3
Full recovery expected for marathon winner
Lexical
3
Flu Vaccine Appears Safe for Young Children
3
News Baby pandas! Baby pandas! Baby pandas!
Table 5: Top negative mood tweets. The top ranked post shows the limitations of a purely lexical approach,
misreading an onomatopoeia for a negative word. The 8th ranked post is also poorly ranked because of
words like ’lost’, ’small, and ’insignificant’. Consider, though, the amount of context that is needed to know
that this is a post about weight loss (good) rather than currency loss (bad) in the U.K.
Score Tweet
8
dun dun dununun dun dun! dun dun dununun dun dun! ”Ghostbusters!”
7
Had to tackle some skinny ass white mother fucker! His stupid ass almost made me miss my
train. I refuse to wait 15 minutes cause you dumb.
7
Is fed up. Bloody no job. Bloody no money. Bloody hospital. Bloody drugs. Bloody boredom.
Bloody buggering bollocks.
6
RT user4193: #dontyouhate thirsty ass bitches/niggaz ughhh go somehwere with ur thirsty ass!!
(Byeeee! Go die in the desert!!!)
6
Might have to skip work tomorrow... sore throat+sneezing+fever= bad news :( no, not swine flu!
My Mum had this same thing a few days ago.
5
hell yeah that shit bomb i got everybody n the hood doin that shit
5
perhaps sgt. crowley should personally reimburse mass. tax payers for his stupidity & wasteful
expenditure of tax dollars.
5
I can has lost another pound and a half! That’s a quarter of a stone, small but not insignificant
amount!
5
tell that bitch she dont want it; ill kick her ass! tryna mess with my steph smh bird!
3.5 Personality
The essays and Facebook posts corpora are labelled by the Big Five personality properties of each author, so
we can build and assess predictive models for each of the Big Five dimensions. We build and test models for
the essays, Youtube, and Facebook corpora. For the first two corpora, this represents a few hundred words,
and so stands in for what a robot could collect over the course of several interactions with a human. The
Facebook corpus, on the other hand, stands in for a single, short interaction for which it might be expected
to be quite difficult to infer personality traits.
Our process is as follows. For each extreme of a personality dimension (e.g. for Extroversion: extrovert,
and introvert) we select as training data documents authored by individuals with that trait.
For example, to predict extroversion versus introversion, we select 100 randomly chosen documents au-
thored by extroverts and 100 authored by introverts. For each set, we construct a document-word matrix
based on the most-frequent 1000 words, and normalize based on the log + digamma process described pre-
viously. This gives us a 100 × 1000 matrix, A, whose rows represent documents, columns represent words,
and entries represent normalized word pseudo-rates. We normalize the columns of this matrix to z-scores,
recording the mean and standard deviation of each column so that we can normalize test records to match.
We compute the singular value decomposition of A as
A ≈ USV
8
Table 6: Top negative mood Facebook posts. The 3rd ranked post is interesting – the poster is obviously
highly positive, but the post contains a large number of negative words and might arguably be considered
to express considerable negativity as well.
Score
Post
1.5812
(continued from above) And rot inside a corpse’s shell The foulest stench is in the air The funk
of forty thousand years And grizzly ghouls from every tomb Are closing in to seal your doom
And though you fight to stay alive Your body starts to shiver For no mere mortal can resist
The evil of the thriller (Into maniacal laugh, in deep echo)
1.4481
Recidivism (n.) - 1. Committing new offenses after being punished for a crime. 2. Chronic
repetition of criminal or other antisocial behavior. Replace criminal with stupid , and I think
I’ve found my problem.
1.1806
I’m ecstatic! I just spoke briefly with *PROPNAME*’s lawyer. Based on some case files
I emailed her, we have enough evidence to prove CPS’s removal of the kids was illegal and
grounds for an emergency hearing. We believe we have enough evidence to cost both CPS
and Humble PD their qualified immunity , allowing us to file a civil suit against them and the
apartments. ANYONE KNOW A GOOD CIVIL ATTORNEY???
1.1303
Fight one more round. When your arms are so tired that you can hardly lift your hands to
come on guard, fight one more round. When your nose is bleeding and your eyes are black and
you are so tired that you wish your opponent would crack you one on the jaw and put you to
sleep, fight one more round remembering that the man who always fights one more round is
never beaten
1.1168
will gladly boast in his weaknesses. When I am weak, then He is strong. Your might, oh Lord is
endless. Please cover over me. Please cover over your beloved..Lest we all fall in overwhelming
shame under the weight of our imperfections.
1.0684
Heh...: God I wish that I could hide away//And find a wall to bang my brains//I’m living
in a fantasy,//a nightmare dream...reality//People ride about all day//In metal boxes made
away//I wish that they would drop the bomb//And kill these cunts//that don’t belong! I hate
people!//I hate the human race//I hate people!//I hate your ugly face//I hate people!//I hate
your fucking mess//I hate people!//They hate me! -Anti-Nowhere League
1.0555
Like fire, Hellfire, This fire in my skin. This burning, desire, Is turning me to sin. — It’s not
my fault; I’m not to blame. It is the gypsy girl; The witch who sent this flame — It’s not my
fault; If in God’s plan. He made the devil so much stronger than a man
1.0539
I have a ship and a fighting crew and a girl with lips like wine, and that’s all I ever asked. Lick
your wounds, bullies, and break out a cask of ale. You’re going to work ship as she never was
worked before. Dance and sing while you buckle to it, damn you! To the devil with empty
seas! We’re bound for waters where the seaports are fat, and the merchant ships are crammed
with plunder!
1.0028
I don’t get some people...you offer to help them- they are lazy and don’t bother to send their
shit... months later when its too damn late they ask you to help them.. wtf
0.99347 Sick of this nihilistic depression shit–fine, life may be meaningless, I may be right about all the
other disturbingly bleak shit I’ve been thinking...but I’m not a self-killer so I’m stuck here–
might as well pull it together and live this crazy life I’ve got anyway, eh? Some change is in
the works (just in time for the New Year, ironically).
where U is an n × k matrix representing the documents in k-dimensional space, S is a diagonal matrix with
non-increasing entries representing the amount of variation in each dimension, and V is a 1000 × k matrix
representing the variation among the words.
Given a test record, expressed as word pseudo-rates, we apply the means and standard deviations used
for the training data to map it to a scale that matches the training data. Then we recompute the SVD
with the test record added to the end of A, compute a new V matrix, ˆV, which is still v × k, and compute
the Frobenius norm difference between V and ˆV. (In fact, we do not need to recompute the SVD; see the
9
Table 7: Top negative mood headlines. Again the human-ranked headlines require more context to recognise
their negativity.
Method
Score Post
-100
Bombers kill shoppers
Human
-98
Mortar assault leaves at least 18 dead
-98
Iraq car bombings kill 22 People, wound more than 60
4
Brazil Air Force Cites Faults and Confusion in Fatal Crash
Lexical
4
Freed kidnap suspect: my terror at police raid
4
Iraqi suicide attack kills two US troops as militants fight purge
Table 8: Extracts from essays scoring high on positivity and/or negativity
Post
Comment
I really don’t want to go back to LOCNAME, I get lonely. I
hate this stupid town, but I don’t want to leave home, but then
again I do , the air is cold when I breathe in and I need to go
take my medicine oh well I’ll go do that later crap I’m tired
and I don’t want to go read either it gets so long and boring
very negative essay
I feel gross because it’s the end of the weekend and I was out
partying the whole time. ... But, I am having so much fun.
essay that is both positive and negative
I started to cry, out of sheer joy at what he was saying to me,
I have never had any one person ever express their feelings to
me
strongly positive essay
appendix.)
If the test records ‘fits’ with the model built from the original training set, then the difference in the
Frobenius norm will tend to be larger than if it does not. This can be seen by considering the properties of
the SVD. One interpretation of an SVD is that, if the rows of U and the rows of V are plotted in the same
k-dimensional space, the distance between a point representing a row of U and a point representing a row
of V corresponds to the affinity between that object and that attribute. Adding a new point corresponding
to an added row of U has a different effect, depending on how similar it is to the existing rows. If it is like
them, then it exerts strong pulls on all of the attribute points because it, like the rest of the object points,
is close to them. If, however, it is not like the existing objects, then it is plotted far from them, and it is
therefore far from the attribute points as well. Hence, it exerts only a weak pull on them; which is reflected
in a small change in norm.
We have no scale that can be used to determine when a norm change is large or small. We use both of
the models from the extremes of each personality dimension, the models for extroverts and introverts. Given
Table 9: Extracts from vlogs scoring high on positivity and negativity
Post
Well, I am so excited to talk to you today about goals, strategy and tactics, because I – this is really
the secret sauce. First off, the goal. Might sound like a no brainer but some people forget what it is, to
increase revenue. Hello, we’re in the business of making money, that’s what we’re supposed to be doing
here. Uh, now a lot of people think of the strategy as being the goal. The strategy I’m going to use is to
increase audience. That’s a strategy to achieve that goal.
Okay, this is part two of my anarchy Q and A session. XXXX says I vote for capitalism, which, like
democracy, is a horribly flawed system that happens to be the best thing we can come up with at the
moment. However, even capitalism is starting to show its flaws like the social ones. I seriously doubt
capitalism will survive another fifty years at current rates of technological development. We’re perilously
close to technologies capable of surpassing the abilities with respect to the specific jobs of most of the
morons in our population.
10
Table 10: Top anticipation tweets. Note the (undetected) sarcasm in the 10th ranked tweet.
Score Tweet
7
RE: High Road - Glad to help. U pray I pray We all pray. God provides. Our God is an awesome
God! keep on keepin on
6
good luck good luck good luck!
5
- morning and happy national day! good nite sleep often lead to a good start for the day. enjoy
the rest of your day!
5
thought of ya the other day. Time will come and just remember all the good stuff ya can. God
bless!
5
hey! just wanted to say ive enjoyed praising god with you at church! You have an amazing gift!
Im glad you are getting to share it
5
yes its as long as U wake up alive it’s enough reason 2 celebrity life & enjoying life, we make R
self happy, happines is not gained
5
LETS GO GIANTS!! ::clap, clap, clap clap clap!:
5
glad you had a good time, thanks for coming. Hope to see you at the next one which should be
in 2 weeks.
5
sure it’s just as good... like white chocolate is just as good as real chocolate.
5
Yh you did start it! Bloody hell, I’m nervous now. Tell you what, I knew Harmison was gonna
enjoy this pitch. Good start!
Table 11: Top anticipation Facebook posts
Score
Post
1.4652
Please keep *PROPNAME*’s family in your prayers tonight and tomorrow. We have to be in
court at 9am. Really pray for the oldest *PROPNAME* went to his dad’s for his summer visit
and had his mother, brother and sister stripped from him for six months. CPS is providing
Christmas for the other two but not him. I sure hope we succeed at stripping them of their
immunity and make them pay for what they’ve done to this family.
1.1556
Dedicate yourself to the good you deserve and desire for yourself. Give yourself peace of mind.
You deserve to be happy. You deserve delight. *PROPNAME*
1.0844
Was on my first voice lesson today... feels good to start singing again... It was a long time
ago... but now I’m back on track again ;)
1.0536
long day of classes//library time... but.... *PROPNAME* is coming to see me tomorrow :) So
everything is good.
1.0536
feels pretty ooky, but is going to try to power through and sing in everything tomorrow anyway.
Good idea? Let’s watch!
1.033
Seventh day of Christmas!!! I still hear the church bells ringing and Handel’s Hallelujah Chorus!!
Glory to God in the highest and on earth peace to all good men!!!
0.97647 has contacts finally, its been a long time coming. its quite nice to see
0.96156 ohh, cleaning then reading for school... all the while just thinking about how much I can’t wait
to go home on Thursday! Lots of time with my one and only, hunting, and all the fun and good
people that go along with both! Can Not Wait.
0.96156 Finally got a car! 1998 Nissan Altima GXE w// 98,700 miles on it in pretty good shape.
Mechanics aside, glad I have a car w// a stereo again. Don’t like that it’s white, but I took
what I could get! Might either mural-paint or shop-repaint it if it annoys me enough. Also had
a really great weekend (Thank you, guys. Xoxo!)–nice change from the last few! Glad as hell
I’m not stranded anymore!!
0.96156 Hard drive problem? Made funny noises twice and failed to boot. Finally booted (obviously).
I am going to install a new one now. Don’t expect updates for a few minutes (hah!). Of course
the last time this happened, I installed a new drive, used the old one for backup, and the old
continued working for years. So this could all be a waste of time and money.
11
Table 12: Top joy tweets
Score Tweet
7
RE: High Road - Glad to help. U pray I pray We all pray. God provides. Our God is an awesome
God! keep on keepin on
6
i am becoming VERY NAUGHTY in good and bad way. Haha. I will just hugg and kiss everybody
and dance dance dance dance : ...
6
good luck good luck good luck!
6
Peter Please Reply Me!! I Think That You Are Helping A Very Special Cause Good Luck And
God Bless You! kiss!
5
well hope its all you hope it is .....a lot of hope i realise but im quietly confident you will have a
wonderful timexxx
5
It won’t be long until you find love, and by love, I mean TRUE love. You are such a beautiful,
sweet, intelligent,
5
love is energy yes - love is everything - the door, the chair, the tree, the house, the wind, God, to
me love is everything
5
hey! just wanted to say ive enjoyed praising god with you at church! You have an amazing gift!
Im glad you are getting to share it
Table 13: Top joy Facebook posts
Score
Post
1.2444
Define me urban dictionary The Most wonderful person in the World. Kind, Sweet, Loving,
Caring, Gentle. Perfect in Every way. The one you love for all your life. Crazy hot girl.
Beautiful, smart & funny; *PROPNAME* posesses atributes absent in 99.9% of women. Truly
a lucky find. Plus she rocks. The most wonderful drug in the world, better know as lortab.
taking the pill may cause a sense of euphoria, & well being. hahaha
1.2319
Let me know you more deeply and truely, oh Lord. There are none who know you completely,
and so I pray that YOU show me. Your ways are perfect and I want my life to sing your praise.
Let me walk with love, joy, peace, patience, kindness, goodness, gentleness, and self-control.
Against such things there is no law.
1.1556
is thankful for everyone and everything. My life is beautiful and I am such a lucky girl. ? &
?...Happy Thanksgiving! Love you all!
1.1484
Happy Thanksgiving everybody! I’m thankful for friends to share Thanksgiving with:) Now
time to make some yummy food to share later
1.1201
is wishing a Happy Birthday to my bestie *PROPNAME*! I love you honey, and I hope you
have a wonderful 22nd Birthday!
1.1201
wishing my fiance a Happy, Happy Birthday!! I love you darlin’ and hope you have a wonderful
day!
1.033
Seventh day of Christmas!!! I still hear the church bells ringing and Handel’s Hallelujah Chorus!!
Glory to God in the highest and on earth peace to all good men!!!
1.033
Wish you all a VERY HAPPY ’Christmas’..enjoy with your family and friend..spread the smile..
Look in your socks in the morning and you ill find a gift - from Him.. Lets spread the Joy nd
pray to let peace reign everywhere in the world..
0.9768
Please keep *PROPNAME*’s family in your prayers tonight and tomorrow. We have to be in
court at 9am. Really pray for the oldest *PROPNAME* went to his dad’s for his summer visit
and had his mother, brother and sister stripped from him for six months. CPS is providing
Christmas for the other two but not him. I sure hope we succeed at stripping them of their
immunity and make them pay for what they’ve done to this family.
0.97647 joy and freedom in the love of God!
12
Table 14: Top joy headlines. The human-ranked headlines do not seem particularly joyful; but nor does the
3rd lexically ranked headline, misled by the ambiguity of ’baby’ and ’found’, but the first lexically ranked
headline is indisputably joyful.
Method
Score Post
93
Goal delight for Sheva
Human
93
Lily Allen wins web music award
91
China Successfully Launches Two Satellites
3
News Baby pandas! Baby pandas! Baby pandas!
Lexical
3
’Jackass’ star marries childhood friend The secrets people reveal
3
Outdated baby food found on shelves
Table 15: Top surprise tweets. These suggest that ’hope’ and ’luck’ are over-associated with surprise in the
lexicon.
Score Tweet
6
good luck good luck good luck!
5
Ok-I am surprisingly zero’d out.time to catch an early one;I have a feeling I’m going to lose this
weekend as well.nite nite all!tweet good!
5
Mosley can suck ma’ balls. Good on FOTA for putting their money where their mouth is. I hope
FIA caves and we get goodness.
4
well hope its all you hope it is .....a lot of hope i realise but im quietly confident you will have a
wonderful timexxx
4
user10 Hiya Moira. Hope you had a lovely weekend. Did you have good weather down south?
Twas OK up here. Hope youre well xx =]
4
hope u dont mind but using your lovely sunset shot of serpentine as screen saver the colours are
lovely u have talent xxx
4
I NEED TO LOSE WEIGHT MAFUCKERRRRRRRRRRRRRRRR.I WANT TO LOSE
WEIGHT.FUCK FOOD FUCK FOOD.FUCK ALL GOOD FUCK.
4
good, good. nahh, im just worried about everyone else. sorry i ruined it for you. i hope this wont
wreck our friendship
4
what’s good young lady?? Hope ur feeling great today...whatcha doing this weekend???
4
Drive a crappy car to deal so they know you have zero money... Then dress in clothes that are
obviously too small. Good luck!
a test point, we predict it to belong to the class (extreme) that claims it most strongly.
Table 36 shows the performance of predictors built in this way for all five personality dimensions, and for
the essays, Youtube, and Facebook datasets. There are no directly comparable results using other techniques,
since these tend also to use other syntactic markers, but most achieve F-scores in the 60% range, while ours
are typically higher, sometime much higher.
There are interesting differences across the dataset. For the essays dataset, prediction accuracy for all five
dimensions is comparable; for the Youtube dataset, prediction of the conscientious dimension is noticeably
higher than the others, but all are higher than for the essays. For the Facebook dataset, prediction accuracy
for openness is higher than for the other dimensions, but all are lower than for the other two datasets. This
is not surprising given that Facebook posts are relatively short; indeed, what is surprising is that aspects of
personality can be predicted from such short texts. We doubt the many humans would feel confident about
doing so.
3.6 Semantic differential
Osgood postulated that there was a natural three-dimensional structure underlying human value judgements,
in which all words could be considered to lie at coordinate positions with respect to three independent
conceptual axes, which he named the semantic differential [17]. He suggested that words varied along
13
Table 16: Top surprise Facebook posts. Again the association of the words ’death’ and perhaps ’wishes’ with
surprise seems too strong; but the 7th ranked post shows that our normalization strategy is effective.
Score
Post
0.82979 finally feeling somewhat moved in. It might be a miracle folks!
0.77958 Practice question for my EMT exam its early afternoon and the alarm sounds You are called to
a scene at a railroad one of the workers is trapped between 2 cars. He is alert and tells you that
as soon as the cars are moved he will die. He asks that you leave him alone with his wife and
to do nothing heroic. which emotional response to death is your patient exhibiting? ...uhhhhh
how intense is that?!
0.72296 The most beautiful experience we can have is the mysterious. It is the fundamental emotion
which stands at the cradle of true art and true science. Whomever does not know it and can
no longer wonder, no longer marvel, is as good as dead, and his eyes are dimmed. -Einstein
0.719
is wishing a Happy Birthday to my bestie *PROPNAME*! I love you honey, and I hope you
have a wonderful 22nd Birthday!
0.69317 thanks for the birthday wishes
0.69317 Nordea DEATH.
0.69317 EARTHQUAKE!!!!
0.69317 My mouth always gets me in trouble...
0.69317 feels like Death warmed over...
0.69317 scored about 65k in bejeweled blitz in one second. That was nice.
Table 17: Top surprise headlines. Neither the human-ranked or lexically-ranked headlines seem particularly
surprising.
Method
Score Post
87
Man rides stationary bike for 85 hours
Human
79
Archaeologists find remains of couple locked in a hug
77
Kathmandu: First snow in 63 years
3
PM: Havana deal a good experiment
Lexical
2
4th person dies after receiving flu shot
2
Hungary police tackle violent protestors
Table 18: Top anger tweets
Score Tweet
7
Is fed up. Bloody no job. Bloody no money. Bloody hospital. Bloody drugs. Bloody boredom.
Bloody buggering bollocks.
5
hell yeah that shit bomb i got everybody n the hood doin that shit
5
u ain’t never lie I seen 3 white men in the ceelo spot smfhhh!! Shit crazy they tryin to rape use
for money land & bitches!! Lmao
4
damn how yall mad when i gave yall 2months to get yall shit together im backkkk lmfaoo my
cocky shit yall like to hear
4
Joe Jackson is a damn DISGRACE. This man is promoting a new record company and his son,
Michael, hasn’t been dead for 2 days. DAMN DISGRACE
4
Yo Bobby Digital is stuck in my head Lol she be like ”Nigga, You Aint Shit..Ya Daddy Aint Shit,
Ya Money Aint Shit” Lol im bout to bump it
4
Ok. 1 vote for LAC, 1 vote for Minny, and 1 vote for Bucks so far. C’mon guys, we need more
NBA hell teams.
4
I hate it when a girl runs from you screaming ”RAPE!” when all I intended to do was eat her
down to a screaming skeleton to gain her power.
14
Table 19: Top anger Facebook posts. The 10th ranked post is an example of mock anger, and the technique
fails to detect the ironic language.
Score
Post
0.94206 Recidivism (n.) - 1. Committing new offenses after being punished for a crime. 2. Chronic
repetition of criminal or other antisocial behavior. Replace criminal with stupid , and I think
I’ve found my problem.
0.91955 s wireless internet keeps spazzing out every five minutes. It’s really annoying. Actually, it’s
more than annoying, but I haven’t gotten angry enough to start using foul language.
0.77039 Intelligence is a curse, life a senseless disease. Gods grant me respite by death or lobotomy...
0.73236 Parent’s Evening + Bomb Scare = Mucho Chaos.
0.71499 is way too upset with the damn Dodgers to even really be upset.
0.71487 Feeling DAMN GOOD headwise lately: forceful self-change of outlook, meds seem to be finally
working, & I’m startin to get shit done! ’Course, meds need tweaking–one of ’em (dunno which)
is also tryin to make me a zombie–not the fun movie kind, either: I get sleepy around 4 or 5
PM now; too damn early even if I weren’t a night owl! PSYCHED bout the forward progress,
but this zombie thing’s crampin’ my nightlife, oy!
0.7
8 pages in 2 hours...not bad. well, really bad actually! extremely bad what the hell am I saying
not bad hahahaaaaaaaaaa
0.69317 and mosquito bites
0.69317 Nordea DEATH.
0.69317 down to 10 men and we’re still kicking arse
Table 20: Top anger headlines. Both sets are plausible, but again the human-ranked list depends on implicit
context rather than content.
Method
Score Post
71
Israeli woman’s tirade spurs PM outrage
Human
70
Abbas condemns Israeli massacre in northern Gaza Strip
70
Anger at release of two held over beheading plot
3
Poison Pill to Swallow: Hawks Hurting After Loss to Vikes
Lexical
3
Brazil Air Force Cites Faults and Confusion in Fatal Crash
3
Iraqi suicide attack kills two US troops as militants fight purge
the dimensions of good–bad, active–passive, and strong–weak (sometimes called Evaluation, Activity, and
Potency). This construct was validated extensively across cultures, beginning with adjectives but then
extended to a range of other words [16]. Osgood’s view of this three-dimensional structure was attitudinal
in flavor: “The way the lexicon carves up the world” (p171). The semantic differential approach gave rise
to a line of research that still continues in marketing, and led indirectly to the popularity of Likert scales
for eliciting opinions. Russell and Mehrabian [22] characterize essentially the same three dimensions as
underlying the full spectrum of emotions, but this is not surprising since Osgood’s claim was that these
three dimensions are “a universal framework underlying certain affective or connotative aspects of language”
(Ibid).
This semantic differential is one conceivable way to begin to model attitudes (and perhaps eventually
values). In our context, the semantic differential scores of an utterance can be considered as capturing the
attitudes of the speaker to the general immediate context. Both tweets and Facebook posts can be considered
as declarative statements of the worldview of the author at the moment that they are uttered. Some have
an obvious target in mind; others are simply general comments on life as perceived by the author. In the
context of Facebook this is explicit in the naming of the post as a ‘status update’.
It is possible to imagine a robot inferring more sophisticated attitudes by computing the current object of
attention of a human in its vicinity (for example, using gaze), and associating attitudinal markers in speech
with this object. This is still an area in its infancy.
We take the same corpora previously used to infer emotions and compute the three intensities of each
15
Table 21: Top disgust tweets. Note the overlap with high-ranked anger tweets, reflecting that some emotions
are not well separated by the empirical approach to constructing the NRC lexicon.
Score Tweet
6
Is fed up. Bloody no job. Bloody no money. Bloody hospital. Bloody drugs. Bloody boredom.
Bloody buggering bollocks.
5
i think its because im sick. Doc got me blowin a balloon with my nose and shit... shit aint workin.
This shit is madd annoying...
4
Hell yea ” So you callin me a liar!?” Shit hell yea!
4
hell yeah that shit bomb i got everybody n the hood doin that shit
4
i never sleep too.. its shit. yeah i been ill since wednesday night, been well bad, i been semi
diagnosed with swine flu
4
damn how yall mad when i gave yall 2months to get yall shit together im backkkk lmfaoo my
cocky shit yall like to hear
4
Joe Jackson is a damn DISGRACE. This man is promoting a new record company and his son,
Michael, hasn’t been dead for 2 days. DAMN DISGRACE
4
ysl is the shit vavy @user3985 nigga i found all the hard yeve saint laurent shit the got shit for
the low lol belts only 275 cop out boy!
4
I hate it when a girl runs from you screaming ”RAPE!” when all I intended to do was eat her
down to a screaming skeleton to gain her power.
document on the three Osgood dimensions. Heise [10] computed loadings on each of the three dimensions
for the most common 1000 words in English. We construct document-word matrices from the documents in
each corpora as before, and weight each word by the corresponding loading for each of the three dimensions.
We then score each document by adding up the weights of each word it contains. The loadings have values
between −1 and +1, so the most intense document in each dimension will have a large positive score, and
the least intense a large negative score.
These scores reflect the attitude of the author to the external world at the instant it was written. So,
for example, ‘good’ reflects the author’s view of his/her context, as opposed to positive mood, which reflects
the author’s internal state. As before, we list the top ranked at each extreme to illustrate. These are shown
in Table 37 (top good tweets), Table 38 (top good Facebook posts), Table 39 (top bad tweets), Table 40
(top bad Facebook posts), Table 41 (top active tweets), Table 42 (top active Facebook posts), Table 43 (top
passive tweets), Table 44 (top passive Facebook posts), Table 45 (top strong tweets), Table 46 (top strong
Facebook posts), Table 47 (top weak tweets), and Table 48 (top weak Facebook posts).
The results from the Osgood model are not as compelling as the results for detecting emotions. We
suspect that this is because tweets and Facebook posts are only attitudinal in quite diffuse ways, that is
they reflect a general attitude to the world at the moment of authoring and so conveyed in words whose
weightings are quite small.
4 Discussion and conclusions
We have considered the problem of inferring multiple aspects of human affective state from the perspective
of a robot which (over)hears the conversation of humans in its vicinity. For inferring rapidly changing
components of affective space, such as emotions, computations must be done from small amounts of text
and in short time frames. For components that change more slowly, greater amounts of text and longer time
frames are plausible.
A robot must understand the emotional state of humans with which it interacts to complete the affective
loop, which seems to be essential to creating truly social interactions. Mood, in turn, must be understood
because of its modulating effect on emotion. The ability to detect wishes, desires, and preferences is the
first step towards a robot that can act without explicit commands in ways that the humans it serves want,
reducing the cognitive burden on those humans. The ability to infer attitudes and personality are useful for
a robot to winnow, in advance, the likely repertoire of decisions and actions a human might take in particular
16
Table 22: Top disgust Facebook posts. The 9th ranked post expresses many emotions, but the disgust score
comes from the use of ‘zombie’ and ‘shit’.
Score
Post
0.79003 worst night ever! then I get this message? hey so i just moved up here and ill be honest, living
with someone but not getting needs met sexually and hoping to find someone in same position
or at least who understands and can be discrete, if you may be interested let me know ,if not
i am very sorry to have bothered you I could be a god damn serial killer who gets off setting
people on fire urg I fucking hate people!
0.77338 Recidivism (n.) - 1. Committing new offenses after being punished for a crime. 2. Chronic
repetition of criminal or other antisocial behavior. Replace criminal with stupid , and I think
I’ve found my problem.
0.77041 Heh...: God I wish that I could hide away//And find a wall to bang my brains//I’m living
in a fantasy,//a nightmare dream...reality//People ride about all day//In metal boxes made
away//I wish that they would drop the bomb//And kill these cunts//that don’t belong! I hate
people!//I hate the human race//I hate people!//I hate your ugly face//I hate people!//I hate
your fucking mess//I hate people!//They hate me! -Anti-Nowhere League
0.77039 Intelligence is a curse, life a senseless disease. Gods grant me respite by death or lobotomy...
0.76996 Well, lived through the Carbon Monoxide shit, though it made me sick today–luckily not sick
enough to go to the damn hospital I guess; just sick enough to be puking and bedridden today.
Good thing I had the day off anyway for MLK day (but my dad didn’t tell me that, so I woke
up ill and panicky this afternoon–nice, lol). Furnace was fixed today, though, so no more CM
troubles, I suppose!
0.7528
I have a ship and a fighting crew and a girl with lips like wine, and that’s all I ever asked. Lick
your wounds, bullies, and break out a cask of ale. You’re going to work ship as she never was
worked before. Dance and sing while you buckle to it, damn you! To the devil with empty
seas! We’re bound for waters where the seaports are fat, and the merchant ships are crammed
with plunder!
0.73236 is tired of being sick. Death to all swine.
0.71499 got sick of the crap 5 haircuts so upgraded to the 22 haircut today.still crap.
0.71487 Feeling DAMN GOOD headwise lately: forceful self-change of outlook, meds seem to be finally
working, & I’m startin to get shit done! ’Course, meds need tweaking–one of ’em (dunno which)
is also tryin to make me a zombie–not the fun movie kind, either: I get sleepy around 4 or 5
PM now; too damn early even if I weren’t a night owl! PSYCHED bout the forward progress,
but this zombie thing’s crampin’ my nightlife, oy!
Table 23: Top disgust headlines. The human-ranked list depends on knowing that teachers, especially, are
not supposed to assault others sexually, and that child porn is especially disgusting.
Method
Score Post
87
Teacher charged with sex assault
Human
74
Abbas condemns Israeli massacre in northern Gaza Strip
73
Investigation of child porn site hits 77 nations
2
Teen repellent inventor wins infamous prize
Lexical
2
Israeli woman’s tirade spurs PM outrage
2
Observatory: poison begets poison
situations, and so allow the robot to precompute likely responses it might be called upon to make.
We use tweets, headlines, and Facebook posts as surrogates for conversations, and essays and vlog tran-
scripts as surrogates for what a robot might accumulate about a human in the course of multiple interactions.
These documents are extremely ungrammatical, containing multiple abbreviations, misspellings, and jargon.
Some of these would remain in conversational snippets, but some would not.
We do not have ground truth for the emotions, wishes, desires, and attitudes of the authors of each
17
Table 24: Top fear tweets. The lexicon struggles with this emotion because tweets tend to contain hyperbole:
‘die’, ‘dying’, and ‘drowning’ are not the fear-laden words in this context that the lexicon assumes they are.
Notice how much social context would be required to understand that these are not fear words in these
tweets.
Score Tweet
7
Is fed up. Bloody no job. Bloody no money. Bloody hospital. Bloody drugs. Bloody boredom.
Bloody buggering bollocks.
6
RE: High Road - Glad to help. U pray I pray We all pray. God provides. Our God is an awesome
God! keep on keepin on
5
It’s Godwin’s Law. It’s also insulting, considering the last 8 years of Bush rule, I doubt these
fuckers were screaming foul.
4
agree problematic as the ”war on terror” has been labeled by some as the endless war
4
This lil girl wants 2 watch spongebob squarepants bitch I’m 20 yrs old I’m grown either we watch
Blues Clues or we won’t be watchin tv @ all
4
its going Great!! Keeps me Honest!! Its fun to look for the best bang for my buck!! Are u gonna
watch BB11 tonight?
4
I hate it when a girl runs from you screaming ”RAPE!” when all I intended to do was eat her
down to a screaming skeleton to gain her power.
4
Chicken kebab from local chipper is to die for. No doubt dying is what I will be in the morning.
4
One thing I don’t mind drowning in is confidence. Hell, I live in a confidence desert. - -URL-
4
I HATE awkward moments in movies. They make me switch tabs and do something else to avoid
the imminent embarrassment. Why do I even bother?
document. We appeal to face validation by showing those documents that rank most highly for each of the
properties considered. In general, it is clear that the high-ranked documents do exhibit the relevant property
but, of course, this does not mean that other documents that also exhibit the property have not been missed.
For personality, we use datasets for which ground truth is known. Even humans typically achieve no better
than 80% accuracy for personality trait prediction. Hence, our algorithm’s performance, achieving prediction
accuracies mostly in the 70% range, is strong.
The surprising conclusion of our analysis is that computational models perform usably well, even from
quite small amounts of text, and with relatively simple analysis. It is therefore plausible that robots could
build models of the affective state of the humans they encounter using practical amounts of computation
time and resources.
Avoiding recomputing the SVD
If ˆA is the matrix with extra (test) rows added, then
ˆV ≈ ˆAU+S−1
where U+ is the Moore-Penrose inverse of U (which isn’t square, let alone invertible) given by U+ =
(U U)−1U . ˆV is m × k, ˆA is m × (n + q), U+ is (n + q) × k, and S−1 is k × k.
References
[1] M.C. Ashton, K. Lee, and L.R. Goldberg. A hierarchical analysis of 1,710 English personality-descriptive
adjectives. Journal of Personality and Social Psychology, 87(5):707–721, 2004.
[2] Joan-Isaac Biel, Vagia Tsiminaki, John Dines, and Daniel Gatica-Perez. Hi youtube!: Personality im-
pressions and verbal content in social video. In Proceedings of the 15th ACM on International conference
on multimodal interaction, pages 119–126. ACM, 2013.
18
Table 25: Top fear Facebook posts. Again, the lexicon labels many words as expressions of fear, when they
are often used hyperbolically. The labelling of ‘pray’ as a fear word is understandable but, as the 9th ranked
post shows, is inappropriate.
Score
Post
1.2175
Heh...: God I wish that I could hide away//And find a wall to bang my brains//I’m living
in a fantasy,//a nightmare dream...reality//People ride about all day//In metal boxes made
away//I wish that they would drop the bomb//And kill these cunts//that don’t belong! I hate
people!//I hate the human race//I hate people!//I hate your ugly face//I hate people!//I hate
your fucking mess//I hate people!//They hate me! -Anti-Nowhere League
1.0062
(continued from above) And rot inside a corpse’s shell The foulest stench is in the air The funk
of forty thousand years And grizzly ghouls from every tomb Are closing in to seal your doom
And though you fight to stay alive Your body starts to shiver For no mere mortal can resist
The evil of the thriller (Into maniacal laugh, in deep echo)
0.96299 Intelligence is a curse, life a senseless disease. Gods grant me respite by death or lobotomy...
0.9037
Many of us spend our whole lives running from feeling with the mistaken belief that you cannot
bear the pain. But you have already borne the pain. What you have not done is feel all you
are beyond the pain. -*PROPNAME*
0.88545 I’m ecstatic! I just spoke briefly with *PROPNAME*’s lawyer. Based on some case files
I emailed her, we have enough evidence to prove CPS’s removal of the kids was illegal and
grounds for an emergency hearing. We believe we have enough evidence to cost both CPS
and Humble PD their qualified immunity , allowing us to file a civil suit against them and the
apartments. ANYONE KNOW A GOOD CIVIL ATTORNEY???
0.88246 Put this as your status if you or somebody you know has suffered the LOSS of a Baby ? The
majority won’t put it on because unlike cancer. Baby Loss is a taboo... Break the Silence ? This
is in memory of all the Angel Babies...gone too soon but NEVER FORGOTTEN ABOUT!!?
Show your support and let these women know they don’t have to grieve alone.
0.86085 wonders, with all this righteous indignation over a terrorist assassination program and change
in tactics from pursuing the Taliban to protecting the population, when did this stop being a
war?
0.8594
I want your love and I want your revenge You and me could write a bad romance I want your
love and All your lovers’ revenge You and me could write a bad romance
0.85341 Last night I went to a conference where the police were asking us to pray for them. Lets pray
for the police and our communities. God moves when his people pray.
0.79003 worst night ever! then I get this message? hey so i just moved up here and ill be honest, living
with someone but not getting needs met sexually and hoping to find someone in same position
or at least who understands and can be discrete, if you may be interested let me know ,if not
i am very sorry to have bothered you I could be a god damn serial killer who gets off setting
people on fire urg I fucking hate people!
Table 26: Top fear headlines. The human-ranked list is better, but again because of knowledge about, for
example, letter bombs as a tool of terrorism.
Method
Score Post
95
New Iraq terror tape calls for abducting foreigners
Human
94
Bombers kill shoppers
92
UK workers on alert for letter bombs
5
Freed kidnap suspect: my terror at police raid
Lexical
4
Brazil Air Force Cites Faults and Confusion in Fatal Crash
4
Iraqi suicide attack kills two US troops as militants fight purge
19
Table 27: Top sadness tweets. The associations of ‘black’, ‘blues’, and ‘nothing’ with sadness explain a
number of these high-ranked tweets, but they are common words for which their other meanings are not sad.
Score Tweet
8
Is fed up. Bloody no job. Bloody no money. Bloody hospital. Bloody drugs. Bloody boredom.
Bloody buggering bollocks.
5
Oh please. Dark romance. I love me some dark romance. Just the kind of news to make my bad
day better :-)
4
You were everything thats bad for me, make no apology. Im crushed black and blue. But you
know id do it all again for you.
4
Ok. 1 vote for LAC, 1 vote for Minny, and 1 vote for Bucks so far. C’mon guys, we need more
NBA hell teams.
4
user4461 just in case u were worried I didn’t die of heat stroke or get kidnapped! Lol
4
Nothing goes on the show without testing by the Black and Blues RT user4734 user4739 do the
black and blues even test the motivator?
4
Is it better to be sick on a day off and lose the day but not have to do anything, or be sick on a
show day and lose the day off? Hmmm
4
*rabid foam kill kill kill* :-)
4
The cake is a lie. -NL-The cake is a lie. -NL-The cake is a lie. -NL-The cake is a lie.
4
user6636 good unfair and bad unfair for me. At Crawford. Much worse hours.
[3] Paul T Costa Jr and Robert R McCrae. Domains and facets: Hierarchical personality assessment using
the revised neo personality inventory. Journal of personality assessment, 64(1):21–50, 1995.
[4] R.J. Davidson, K.R. Sherer, and H.H. Goldsmith, editors. Handbook of Affective Sciences. Oxford
University Press, 2009.
[5] J.J.A. Denissen and Lars Penke. Motivational individual reaction norms underlying the Five-Factor
model of personality: First steps towards a theory-based conceptual framework. Journal of Research in
Personality, 42:1285–1302, 2008.
[6] P. Ekman. Basic emotions. In T. Dalgleish and M. Power, editors, Handbook of Cognition and Emotion,
pages 45–60. John Wiley and Sons, 1999.
[7] G. Farnadi, G. Sitaraman, S. Sushmita, F. Celli, M. Kosinski, D. Stillwell, S. Davalos, M.-F. Moens, and
M. De Cock. Computational personality recognition in social media. User Modeling and User-Adapted
Interaction, 26:109–142, 2016.
[8] Golnoosh Farnadi, Susana Zoghbi, Marie-Francine Moens, and Martine De Cock. Recognising per-
sonality traits using Facebook status updates. In Proceedings of the workshop on computational per-
sonality recognition (WCPR13) at the 7th international AAAI conference on weblogs and social media
(ICWSM13). AAAI, 2013.
[9] A.B. Goldberg, N. Fillmore, D. Andrzejewski, Z. Xu, B. Gibson, and X. Zhu. May all your wishes come
true: A study of wishes and how to recognize them. In North American Chapter of the Association for
Computational Linguistics - Human Language Technologies (NAACL HLT), 2009.
[10] D.R. Heise. Semantic differential profiles for 1,000 most frequent English words. Psychological Mono-
graphs: General and Applied, 79(8):1–31, 1965.
[11] Kristina Höök. Affective loop experiences: Designing for interactional embodiment. Philosophical
Transactions: Biological Sciences, 364(1535):3585–3595, 2009.
[12] R. Kirby, J. Forlizzi, and R. Simmons. Affective social robots. Robotics and Autonomous Systems,
58:322–332, 2010.
20
Table 28: Top sadness Facebook posts. For longer documents, the sadness lexicon seems to perform better.
Score
Post
0.9037
Attention everyone: If you have the misfortune of reading this, then I regret to inform you that
I’ve been seriously injured in a car accident. Okay, maybe not, but I did just lose the game....
0.86247 (continued from above) And rot inside a corpse’s shell The foulest stench is in the air The funk
of forty thousand years And grizzly ghouls from every tomb Are closing in to seal your doom
And though you fight to stay alive Your body starts to shiver For no mere mortal can resist
The evil of the thriller (Into maniacal laugh, in deep echo)
0.84339 Recidivism (n.) - 1. Committing new offenses after being punished for a crime. 2. Chronic
repetition of criminal or other antisocial behavior. Replace criminal with stupid , and I think
I’ve found my problem.
0.82789 Sick of this nihilistic depression shit–fine, life may be meaningless, I may be right about all the
other disturbingly bleak shit I’ve been thinking...but I’m not a self-killer so I’m stuck here–
might as well pull it together and live this crazy life I’ve got anyway, eh? Some change is in
the works (just in time for the New Year, ironically).
0.77041 Heh...: God I wish that I could hide away//And find a wall to bang my brains//I’m living
in a fantasy,//a nightmare dream...reality//People ride about all day//In metal boxes made
away//I wish that they would drop the bomb//And kill these cunts//that don’t belong! I hate
people!//I hate the human race//I hate people!//I hate your ugly face//I hate people!//I hate
your fucking mess//I hate people!//They hate me! -Anti-Nowhere League
0.77039 Intelligence is a curse, life a senseless disease. Gods grant me respite by death or lobotomy...
0.75543 Hmm, that wasn’t too bad. Had me worried for a minute. I’ve certainly seen worse endings.
Like freakin’ Fallen! Seriously, what the crap was that?! Anyways, not bad.
0.74509 Gotta love 3 mobile. I asked to cancel my contract and was asked could we convince you to
stay I replied with Due to the terrible customer service, I would rather die a slow and horribly
painful death than stay to which I got do you have any friends or relatives you would like
to refer so they can get a good deal . I want that CSR to work in my company retention
department!!
0.74453 will gladly boast in his weaknesses. When I am weak, then He is strong. Your might, oh Lord is
endless. Please cover over me. Please cover over your beloved..Lest we all fall in overwhelming
shame under the weight of our imperfections.
0.73236 ....lost my bromley id. already. curse my absent mind.
Table 29: Top sadness headlines. The human-ranked list is better; all three of the lexically-ranked list fail
to grasp implicit content.
Method
Score Post
96
Iraqi death toll exceeded 34,000 in 2006, United Nations says
Human
95
Iraq car bombings kill 22 People, wound more than 60
94
Participant in water-drinking contest dies
3
Poison Pill to Swallow: Hawks Hurting After Loss to Vikes
Lexical
2
Injured Marathon Winner Leaves Hospital
2
Beating poverty in a small way
Table 30: Wish lexicon
again, buy, could, find, get, getting, going, hope, hoped, hopefully, hoping, lack, like, liked, looking, love,
need, now, own, please, pleases, should, take, use, want, wanted, wish, would
21
Table 31: Top tweets embodying wishes. This lexicon does seem to perform reasonably at detecting wishes,
even ones that are quite implicit.
Score Tweet
5
What’s your favorite love song? I need to find good ones to download! I need 2 feel the love!
5
religion promotes alot of hate and does more harm then good sometimes! Do want you want,
with who you want, how you want!
5
omg love penry-jones!! my fav austen adaptation - have watched over 50 times. love love love. so
much love.
5
I know u want ME!.. i KNOW u want meee u know i want that XD! ... user11735 man are u
crazy or what!?.. I love to have sex.. i dont want SIDA!
5
What do you like better tweed or spaz I’m thinking I like tweed better I’m the only one in my
circle with the pre need dif view
5
I want a free iPhone so bad I can taste it. ...it tastes less like an Apple and more like plastic and
desperation.#HGBW #FreeiPhone3GS
4
May love surround you and those you love! Be the love that you are and all will unfold better
than you can imgaine!
4
You guys were amazing! I love the new songs! I love you more and more after each show! :-)
4
Hit the road, Wes, and dontcha come back NO MORE NO MORE NO MORE NO MORE!
#Bachelorette
4
*hugs* do hope you fall asleep quickly - sounds like you need more rest. Hope you feel better
soon. : )
4
I, too, love the new vlog :) Keep ’em comin’!! You are both so adorable and I want more too! :D
Love ya!!!
4
Less like a date, more like spending time with an old friend.
4
Oh I don’t think it’s gross, it’s more like I want to high-five him. ’Sides, ain’t like it was uncommon
at the time.
4
so sorry but i love you... (love you more, more)
4
user7808 i’d rather watch will and grace. HA! i want primo, i want DVDs, i want everything.
4
Shaft- do you want the ld or do you want to be held? Random woman- i want the ld then i want
to be held.
[13] Z. Kowalczuk and M. Czubenko. Computational approaches to modeling artificial emotion – an overview
of the proposed solutions. Frontiers in Robotics and AI, 3:21, 2016.
[14] H. Liu and P. Maes. What would they think? a computational model of attitudes. In Proceedings of
the ACM International Conference on Intelligent User Interfaces, IUI 2004, pages 38–45, 2004.
[15] Saif M. Mohammad and Peter D. Turney. Crowdsourcing a word-emotion association lexicon. Compu-
tational Intelligence, 29(3):436–465, 2013.
[16] C.E. Osgood. Semantic differential technique in the comparative study of cultures. American Anthro-
pologist, 66(3):171–200, 1964.
[17] C.E. Osgood, G.J. Suci, and P.H. Tannenbaum. The measurement of meaning. University of Illinois
Press, 1957.
[18] James W Pennebaker and Laura A King. Linguistic styles: language use as an individual difference.
Journal of personality and social psychology, 77(6):1296, 1999.
[19] R. Plutchik. The nature of emotions. American Scientist, 89(4):344–350, 2001).
[20] L. Polyani and A. Zaenen. Contextual valence shifters. In J.G. Shanahan, Y. Qu, and J. Wiebe, editors,
Computing Attitude and Affect in Text: Theory and Applications. Springer, 2006.
22
Table 32: Top Facebook posts embodying wishes
Score
Post
1.3207 Gotta love 3 mobile. I asked to cancel my contract and was asked could we convince you to stay
I replied with Due to the terrible customer service, I would rather die a slow and horribly painful
death than stay to which I got do you have any friends or relatives you would like to refer so
they can get a good deal . I want that CSR to work in my company retention department!!
1.2643 is making an attempt to brief a few cases while waiting for *PROPNAME* to get off work so
she can talk to him... I think I’m actually going to like law school, but I wish my love could be
here with me! I miss him :(
1.1769 Fucking CARBON MONOXIDE alarm’s going off repeatedly again tonight! Opened the windows
& garage doors; I have nowhere else I can go & I need sleep, so I’m gonna hafta just hope the
ventilation is enough cos I don’t know what else to do. Hope I don’t end up in the hospital over
this shit–I’m sick of the damn hospital! (Well...bein’ dead would [debatably] suck too. But I
wouldn’t be self-aware to know about it, lol.)
1.1769 Really? Two days in a row? I would like to take back my comment about de ja vu - I didn’t
mean I wanted it to end like that (again).
1.1218 How messy my room is again.. i love searching for random thngs.. and how nice bed is looking
right now! Ninite all, hope you had great dreams and better sleeps!
1.1218 worst night ever! then I get this message? hey so i just moved up here and ill be honest, living
with someone but not getting needs met sexually and hoping to find someone in same position
or at least who understands and can be discrete, if you may be interested let me know ,if not
i am very sorry to have bothered you I could be a god damn serial killer who gets off setting
people on fire urg I fucking hate people!
1.0894 Dear God, I know you’ve got a lot going on with Jesus’ birthday coming up and all, but I need
you to take this Civil Procedure exam for me in the morning. I know you take all of my exams
for me, but I really need ya in an extra special sort of way for this one. Oh and one more thing,
please hold off the snow until I can make it home....
1.0894 If I should stay, I would only be in your way. So I’ll go, but I know I’ll think of you ev’ry step
of the way...I hope life treats you kind. And I hope you have all you dreamed of. And I wish
you joy and happiness but above this I wish you love....And I
1.0536 thinks that no one should be afraid to go to the doctor because they can’t afford it, no one
should go broke because they get sick, and no one should die because they could not afford care.
If you agree, please post this as your status for the rest of the day.
1.0536 wishes life weren’t so gray-scale... There are things I fear that excite me, things I want that I’m
not sure I want, things I’m doing that I wish I were not, things I should be doing that I dread,
and things that I want to put off but can’t wait to get over with. WTH life, how about giving
me 2+2 instead of [x3-x2]//[(x2+9)3dx]?
1.0536 Well, see what you want to see, you should see it all. Well, take what you want from me, you
deserve it all. Nine times out of ten our hearts just get dissolved. Well, I want a better place or
just a better way to fall.
[21] J.A. Russell. A circumplex model of affect. Journal of Personality and Social Psychology, 39(6):1161–
1178, 1980.
[22] J.A. Russell and A. Mehrabian. Evidence for a three-factor theory of emotions. Journal of Research in
Personality, 11:273–294, 1977.
[23] H.A. Schwartz, J.C. Eichstaedt, M.L. Kern, L. Dziurzynski, and S.M. Ramones. Personality, gender,
and age in the language of social media: The Open-Vocabulary approach. PLoS ONE, 8(9), 2013).
[24] D.B. Skillicorn and P.K. Chandrasekaran. Ranking documents by internal variability. In IEEE Confer-
ence on Intelligence and Security Informatics, Washington, Jun 2012, pages 180–182, 2012.
23
Table 33: Preference/Desire lexicon
like, dislike, love, want, need, prefer, rather, better, worse, more, less, favourite, favorite
Table 34: Top tweets expressing preferences
Score Tweet
5
What’s your favorite love song? I need to find good ones to download! I need 2 feel the love!
5
religion promotes alot of hate and does more harm then good sometimes! Do want you want,
with who you want, how you want!
5
omg love penry-jones!! my fav austen adaptation - have watched over 50 times. love love love. so
much love.
5
I know u want ME!.. i KNOW u want meee u know i want that XD! ... @user11735 man are u
crazy or what!?.. I love to have sex.. i dont want SIDA!
5
What do you like better tweed or spaz I’m thinking I like tweed better I’m the only one in my
circle with the pre need dif view
5
I want a free iPhone so bad I can taste it. ...it tastes less like an Apple and more like plastic and
desperation.#HGBW #FreeiPhone3GS
4
May love surround you and those you love! Be the love that you are and all will unfold better
than you can imgaine!
4
You guys were amazing! I love the new songs! I love you more and more after each show! :-)
4
Hit the road, Wes, and dontcha come back NO MORE NO MORE NO MORE NO MORE!
#Bachelorette
4
*hugs* do hope you fall asleep quickly - sounds like you need more rest. Hope you feel better
soon. : )
[25] C. Strapparava and R. Mihalcea. Semeval-2007 task 14: Affective text. In Proceedings of the 4th
International Workshop on Semantic Evaluations, pages 70–74, 2007.
24
Table 35: Top Facebook posts expressing preferences
Score
Post
0.97647 Gotta love 3 mobile. I asked to cancel my contract and was asked could we convince you to
stay I replied with Due to the terrible customer service, I would rather die a slow and horribly
painful death than stay to which I got do you have any friends or relatives you would like
to refer so they can get a good deal . I want that CSR to work in my company retention
department!!
0.97647 I need to tell people I’m single more often. Two girls talk to me in less than 24 hours. That’s
like twice as many in the last week. Wow. I feel special. That’s why I ride the short bus.
0.83631 My love, I forgive you; you never planned to die - and love, I’ll place two pennies over your
eyes... And I will love you, after the war. I love you for always, forever more. I will love you,
after the war - forever, for always, and more...
0.82979 is going to need a lot more upper body strength if I ever want to be a pole dancer
0.82979 Going off to kick my own ass at the gym once more.... For better or worse, this type of
masochism is highly addictive
0.82979 The day you wake up and realize you’ve just slept twelve hours and still want to sleep some
more is the day you need to get back on a regular sleeping schedule.
0.82979 is being conformed to the image of Christ. Do you want to be like Jesus? Then you must learn
something...you must learn how to love someone who does not meet the conditions. It’s not
easy, but it’s not impossible.
0.82979 Let me know you more deeply and truely, oh Lord. There are none who know you completely,
and so I pray that YOU show me. Your ways are perfect and I want my life to sing your praise.
Let me walk with love, joy, peace, patience, kindness, goodness, gentleness, and self-control.
Against such things there is no law.
0.82979 ...my doctor says I have a cute virus or something like that Bahahahaha No H1N1 here yet!
Bath, lots of liquids and I might even take a nap. Need to get better for date night with
*PROPNAME* tomorrow!
0.82979 Dear Online Dictionary Makers, Let people search the text of the entries, not just the entry
words. Make your dictionaries more like a human brain. Let me search make things the
same and find words that include those words in their definitions. That would rock. Love,
*PROPNAME*
Table 36: Performance predicting Big Five personality dimensions
Dataset/Personality
dimension
Score
Open-
ness
Conscientious-
ness
Extraver-
sion
Agreeable-
ness
Neuroti-
cism
Essays
Av.
0.71
0.69
0.71
0.70
0.70
F1
0.73
0.72
0.72
0.71
0.72
Youtube
Av.
0.79
0.92
0.85
0.79
0.80
F1
0.81
0.92
0.85
0.80
0.80
Facebook
Av.
0.79
0.65
0.67
0.63
0.73
F1
0.82
0.72
0.74
0.70
0.76
25
Table 37: Top good attitude tweets. Because the lexicon is large, and each word has a weighting between
−1 and +1, it is harder to see why each document scores as it does. However, only the 3rd ranked tweet is
hard to understand.
Score Post
7.00
Good Morning!! Come here and I will give you something to do! Really fast!! haha. I’m here
until 2 pm sharp!! :D
6.52
That’s good because we only wear all black to express our inner angst.
6.43
Much more history here. I’m staying at the pub Admiral Nelson was at when he was called up
to fight in what became the battle of Trafalgar.
6.31
dragon ball z was that much more DOPE... back in the daay. narutos not really my thing...
ironing ? good boy : )
5.74
Twitter-ville, you got a good idea for a title for my Special/DVD/CD? I got nothing I really like
and need one tomorrow.
5.73
Car nice and clean for later on! Got my tent fot T! HAd it up and down, packed away in 10 mins!
:) How is every1? x
5.65
yeah poor little baby rosco will have to be in the way back all by himself u need to light a fire
under my butt about my blog lol
5.57
I will be missing the Bayview Social as I will be volunteering at the Wheelchair stand at Festa
Italiana! Got to give back! Enjoy!
5.52
thank u so much user3965 for always being such a good friend & tweeting about the show i’m
producing user4611 w user4612 Love u Kim!
5.40
Only to u... I hav been chatting over there no problem, I must really do some study cause hav
not done any since b4 Melbourne
26
Table 38: Top good attitude Facebook posts. These seem more puzzling than the corresponding tweets,
especially the 4th ranked post.
Score
Post
1.8313 ought to be cast in the next big zombie movie as an extra. Because she feels like the living dead
must feel after a night of heavy brain-eating. Is that their equivalent of drinking? Explains why
they want to do it all the time.
1.803
Out of the Main Event... my 11k didn’t make it far into Day 2. I’ve been playing cash games
non-stop for the last two days though. Probably going to play again today... Now that I’m out,
planning to come to *PROPNAME*’s wedding
1.7976 Prayers needed for 12yr old boy who, after a serious car crash, is now in a critical coma. Pls.
change your status for 1 hr so more people can become aware and add to the prayers. We would
do it for your son or relative, pls. do this for somebody else’s son. Thank you. Pls copy this
and paste..
1.6902 Dear car battery: Why are you dead again? What am I going to do with you?? (No, really–what
am I supposed to do with it?) You make me sad. ://
1.6326 has walked a thousand miles with grace so far. I could walk like this forever...Jesus how can you
be so good to me?? May you be praised in every step I take. Always
1.5979 I believe it all is coming to an end. Oh well, I guess we’re gonna pretend . Ever feel like a
five-year-old who is content to cover his//her eyes to feel safe and secure? If I can’t see you, you
can’t see me..lol..I want us all to stop ignoring the important things. They won’t go away if you
ignore them. I promise.
1.5319 is impressed that the ads on the side of FB are getting relevant without being offensively
so...finally. I’ve got an ad for a book I like, a band I like, and a movie by an animator I
like. Yay!
1.5319 RUDY IS COMING AGAIN!!!
1.5319 Oh, and the 7th of January I’m going for my restricted license... again. Yay.
1.5319 After a slight rebuild... Betsy 2.0 lives to surf the net again!!
Table 39: Top bad attitude tweets
Score Post
-8.60
Well we’re getting a fair amount of the sunny stuff but if you’re out at 2 its probably hard to see.
R u waitin for a sunrise?
-8.60
Cholos r neat,refuse 2 let the style die.where they still get cholo polos at?How does Nike n Reebok
know 2 make the shoes just cholo right?
-8.55
I CAN READ I JUST SEEN U SAY U WAS PUTIN HIM 2 SLEEP IM GON LEAVE U ALONE
FOR RIGHT NOW CUZ I FEEL U MAD AT ME
-8.44
well 1. you should try and pee at the latest 30 mins -1 hr after sex that helps get rid of most of
it then you know just proper
-8.31
you should be greatful of the sun... When you come to Dublin with Miley you will want some
sun... We never get sun.. Just rain
-8.24
So I’m just sitting at home on my own watching rubbish TV...is it wrong to go to bed when it’s
still slightly light outside?Or am I just old
-8.24
Yeah, seriously. Then I just tried again and everything was fine. Still feel like I’ve been put in
my place though. :)
-8.13
user15380 I’m probably gonna get it onthe PS3 when I get paid Wednesday. I can even put my
pic inthe action. Do it. I’ll lend u my PSeye.
-8.02
take a picture of me =P that way u can put it ina frame by your bed and u can smile at me b4
bed everynite, x.x i sound concited
-8.00
I’m throwing out the first pitch at the Twins/Sox game tonight! What should I throw: my
change, fastball, or hook?
27
Table 40: Top bad attitude Facebook posts
Score
Post
-2.4469 [13 hours in vcpa today... it’s going to be a long day..!]
-2.4287 ...when will the GSS end? is BOSE headphone part of the GSS? wanna get one with my
pay...when i get my pay...
-2.4202 I declare myself a mercenary. Have a job? I’ll get it done.
-2.4057 During a recent password audit, it was found that a blonde was using the following password:
MickeyMinniePlutoHueyLouieDeweyDonaldGoofySacramento. When asked why such a long
password, she said she was told that it had to be at least 8 characters long and include at least
one capital.
-2.3984 finished Pride & Prejudice & Zombies instead of watching Ghost in the Shell. I’m classy. Also,
a giant geek. :)
-2.3984 is a Fool but also a Dreamer
-2.3875 will be completely honest for the next 24 hours....you can ask me one question (only in my
inbox)....any question no matter how crazy or wrong it is....but you can only ask me one ques-
tion!!!! you will have my full honesty....but i dare you to put this as your status too and see
what questions you get !!! :) x This should be interesting.
-2.3221 needs to get hair recolored. ahhhhhhhhhhhh
-2.3221 I wanna get two bunnies....They are sooo cute....
Table 41: Top active attitude tweets. It is striking how many of these tweets focus on the issue of getting
up in the morning.
Score Post
7.38
I’m goooood!!! Might go to bed soon with a book! :D early night seeing as I need to get up early
tomorrow! :’( xxxx
6.88
Yes, well perhaps the very short preface of mine. The rest of the book is just pages of ”ouch”
with accompanying drawings.
6.88
I just have it lying around so you may as well have it. About Offspring, have you seen their new
video? Not very good :(
6.82
OKAY! I need to get outta bed now to go to eat before I die of um this nua-ness. So much to do
today! And I need a Fuji ISO 400 film.
6.65
Well, I’m already up, damned world. But I may be as useless as a chimp who can’t throw
bananas...or some better metaphor
6.39
That’s right, I almost forgot. well, have a good day. 4x10’s aren’t so bad, especially when you
have a 3 day weekend ;)
6.23
you could ask them if you could get it switched, call them. But you will prob have to get a new
one. Need help w/ new name?
6.14
OK, I seriously may need someone to start MAKING my ass get up in the mornings....wish I
could make Charlie get me up!!!! LOL...
28
Table 42: Top active attitude Facebook posts. In contrast to the tweets, these posts tend to focus on exercise.
Score
Post
1.9414 just received an extra 360 kb of storage for his computer in the form of a 5 1//4 floppy from the
guy near me who’s retiring. It’s going to come in handy for that extra 20 seconds of music that
won’t fit on my hard drive.
1.8438 [13 hours in vcpa today... it’s going to be a long day..!]
1.8311 long day full of yays and mehs...
1.807
Fucking CARBON MONOXIDE alarm’s going off repeatedly again tonight! Opened the windows
& garage doors; I have nowhere else I can go & I need sleep, so I’m gonna hafta just hope the
ventilation is enough cos I don’t know what else to do. Hope I don’t end up in the hospital over
this shit–I’m sick of the damn hospital! (Well...bein’ dead would [debatably] suck too. But I
wouldn’t be self-aware to know about it, lol.)
1.7347 thinks people should either do something or shut up and deal! We were the (pretty much) first
country to have a peaceful transfer of power between political parties. Let’s go back to those
times, shall we? If you’re not gonna do anything about what you don’t like, then please, shut
up and leave everyone else alone. kthxbi.
1.6869 going for a long run hit me up later
1.6844 gym=the hurt
1.6707 In my shoes, just to see, what it’s like to be me, I’ll be you, let’s trade shoes Just to see what
it’d be like to feel your pain, you feel mine, go inside each others minds Just to see what we’d
find, look at shit through each others eyes
1.6683 Most people never run far enough on their first wind to find out they’ve got a second
1.6437 Hurt my leg today. Might have a stress fracture not sure yet. I can walk on my leg but it HURTS
lol anyways cant wait to see the volley ball game! The Halloween party is still on tonight at
my house. I live like 10 mins away from chief leschi so transportation shouldnt be that big of a
problem
Table 43: Top passive attitude tweets
Score Post
-6.93
it’s hard to believe that’s the same girl that was rockin flour all over her face last year!!! :-D
-6,76
Postcard from Paradise.-NL-You know how hard it is staying out of the sun here. Eclipse re-
hearsals on a week -URL-
-6.61
The dog got 2nd prize at a talent thing at local pet shop - only 2 entrants :) so wasn’t hard :)
-6.48
aww man who you tellin...I tried hard to talk myself outta starting, now I’m tryna talk myself
out of stopping...smh...lol
-6.42
ysl is the shit vavy user3985 nigga i found all the hard yeve saint laurent shit the got shit for the
low lol belts only 275 cop out boy!
-6.37
Grindin hard all day true style of a hustler...not 2 many females out here doin it like that....Well
I must say I am one and will remain
-6.27
invader zim : if you do not know it, we must, must, must provide you with data ...
-6.17
Igor will start the STEPS soon!!! Be ready!!!! Do not miss out this time! Too many fans said
they regret not getting twisted! No regrets!
-6.16
user950 hope you do that quickly! come back soon!
-6.12
Visited bro in hospital. Got a roommate @ the last minute. 80+ year old man who made sure to
tell the nurse he needs a commode. Poor bro.
29
Table 44: Top passive attitude Facebook posts. It is striking that these are all short, that is they use only
a few words but these are strongly passive.
Score
Post
-2.8905 scored another interview!!!!!!
-2.8905 earned another freckle or two from running the *PROPNAME* Marathon today!!
-2.8905 accidentally rescued another animal from T&D’s...
-2.8905 needs another weekend
-2.2181 crowd surfed at *PROPNAME* and *PROPNAME*! =D
-2.1696 is going to the lone star ruby conf. in August!
-2.0909 And there goes another one!!!!
-2.0467 Today is another day. You dont get better by making a right choice once. You get better by
making the right choice... everyday... over and over again...
-1.9894 is drowning in a sea of irrelevance.
-1.7607 is seeing *PROPNAME* this weekend. :) We’re going to the Shedd Aquarium and doing other
fun stuffs. ¡3
Table 45: Top strong attitude tweets. It is striking how much these focus on issues related to sleep.
Score Post
8.58
Good Night Good Morning, hope you had a peaceful sleep and some well needed rest
8.18
Good Morning Tweeps! Morning user7234 - get brekky yet? Welcome back user7228 - get some
sleep, good to have you back!
8.16
Sleep is very good. I went into a deep sleep for the first time in a month.
7.28
yeah I do get a headache when I sleep more than 8 hours and when I sleep when I’m angry
2.....and on most Thursdays
7.20
Good morning! How did you sleep? Did you shake off the undead?
7.12
yep @user14449 me too I feel like hols have begun! Hehe I wish! Yep jus the usual - sleep and
catch ups! We must catch up with Nat as well! xo
6.66
”are you watching me sleep walter?” ”yes your beautiful when you sleep” ”okay watch me sleep
again” ”kay shut your eyes, OMG its BEAUTIFUL!”
6.63
I must sleep again. Good night! xD
6.62
good morning or afternoon, or night lol its morning here, :) sleepy gotta get more then 4 hours
sleep a night lol how r u?
6.52
Describe to me what a good night of sleep is like. I really can’t recall...
30
Table 46: Top strong attitude Facebook posts
Score
Post
3.1886 amber decided it would be fun to steal an egg and smash it on the carpet ):
2.0967 wants to live where soul meets body, and let the sun wrap its arms around her, and bathe her
skin in water cool and cleansing, and feel, feel what it’s like to be new.
2.0482 hmm, can’t wait to try out this synthetic fermented egg lure...
1.8823 is reading Act like a lady, think like a man on her lunch...understanding the male psyche is the
first step to taking over the world :oP
1.8577 is drowning in a sea of irrelevance.
1.7469 just doesn’t feel like working... how about sleep!? :)
1.7082 just heard a commercial on the radio about some pill to lower cholesterol. They actually said
Do you have high cholesterol due to poor diet or lack of exercise? Then you need to try this No,
actually you need to try improving your diet and exercising a bit first. Just a crazy thought I
had...
1.6636 is quite liking girls aloud
1.6636 mmmm beeen feeling quite happpy =))))
1.6468 Nothing quite like going to the park and seeing a vulture munch on a gutted squirrel, turn to
look at you, and precede to empty its bowels with a loud *squirt*
Table 47: Top weak attitude tweets
Score
Post
-10.61 yeah poor little baby rosco will have to be in the way back all by himself u need to light a fire
under my butt about my blog lol
-9.39
Pt 2: Face Place for facials-Tony and Paul are the best hands down in the world & I have been
2 the best places all over the world 2 compare
-9.35
I need u all 2 send positive weather thoughts my way. Fundraising BBQ 2nite and its been
tipping it down all day here. Typical.
-8.22
user6068 user6069 OK guys... just checked nkotb.com ... all the threads are about not being
able to book. So WHO is on the boat then???
-7.89
Le sigh. Thing is, I have a lot of planned posts right up here in my head. But then this is just
so much easier. Wahhh.
-7.77
Fight Club is one of my favorite movies; so much so, it got me to read. And I don’t read much
beyond a few pages of books
-7.52
All he done was show the bad points about Michael and try to make him out as a dick. I hate
Martin Bashir so much.
-7.45
Sometimes what we need and are looking for is right under our nose all along.
-7.31
I want to win something in this contest since I’m only 12 (no income here, only pocket money)
and need iTunes money to buy #simplytweet!
-7.30
You’re obviously not thinking about having fun all day. Sometimes you just need a smile on your
face and everything works out.
31
Table 48: Top weak attitude Facebook posts. It is striking how short they are; again, few words, intensely
weak.
Score
Post
-2.44
my ankle hurts. stupid dishwasher door
-2.2181 6k+car break down+friday the 13th=me feeling superstious. 6k=fail=me tired
-2.2181 will be in knoxville for break. ¿.¡’
-2.2181 working until 1:00pm...really needs a break
-2.2181 IT’S CHRISTMAS BREAK!!!!!!!!!!
-2.1072 Kindergarten report cards
-2.1072 is making his report.
-2.1072 lingaw ako report sa 109. yey :p
-2.031
Today is my 2 month birthday!
-1.9617 work 12-7...
32

Leave a Reply