Social Sentiment Indices Powered by X-Scores
Brian Davis∗, Keith Cortis†, Laurentiu Vasiliu‡, Adamantios Koumpis†, Ross McDermott∗ and Siegfried Handschuh†
∗ INSIGHT Centre for Data Analytics, NUI Galway, Ireland
Email: name. surname@ insight-centre. org
† Universität Passau, Passau, Germany
Email: name. surname@ uni-passau. de
‡ Peracton, Dublin, Ireland
Email: laurentiu. vasiliu@ peracton. com
Abstract—Social Sentiment Indices powered by X-Scores (SSIX)
seeks to address the challenge of extracting relevant and valuable
financial signals in a cross-lingual fashion from the vast variety
of and increasingly influential social media services, such as
Twitter, Google+, Facebook, StockTwits and LinkedIn, and in
conjunction with the most reliable and authoritative newswires,
online newspapers, financial news networks, trade publications
and blogs. A statistical framework of qualitative and quantitative
parameters called X-Scores will power SSIX. This framework
will interpret financially significant sentiment signals that are
disseminated in the social ecosystem. Using X-Scores, SSIX
will create commercially viable and exploitable social sentiment
indices, regardless of language, locale and data format. SSIX and
X-Scores will support research and investment decision making
for European SMEs, enabling end users to analyse and leverage
real-time social media sentiment data in their domain, creating
innovative products and services to support revenue growth with
focus on increased alpha generation for investment portfolios.
Keywords–social sentiment index; cross-lingual; social media
analytics; sentiment analysis; big social and news data.
I. INTRODUCTION
The emerging use of social media data as part of the
investment process has seen a rapid increase in uptake in
recent years, as by examined Greenfield [1]. The lag be-
tween Social Media Monitoring and Social Media Analytics-
“Brand Analytics” and Finance specific analytics applications
has narrowed. The Social Finance Analytics sector has built
on the base developed by Brand Analytics and has evolved
the ecosystem to focus on investment decision-making. The
growth of trading specific social networks like StockTwits
has also provided highly valuable structured social data on
trading discussions, which was not accessible previously on
general social media communities. This new data source has
provided a vital pipeline of thoughts, words and decisions
between people; connecting and interacting as never before.
This collective pulse of conversations and emotional attitudes
acts as a gauge of opinions and ideas on every aspect of
society. Finance specific social media applications provide
asset managers, equity analysts and high frequency traders with
the ability to research and evaluate subtle real-time signals,
such as sentiment volatility changes, discovery of breaking
news and macroeconomic trend analysis. These data streams
can be incorporated into current operating models as additional
attributes for executing investment decision-making, with a
goal to increase alpha and manage risk for a portfolio.
The European research project Social Sentiment Indices
powered by X-Scores (SSIX-http://ssix-project. eu/), seeks to
assist in this challenge of incorporating relevant and valuable
social media sentiment data into investment decision making
by enabling X-Scores metrics and SSIX indices to act as
valid indicators that will help produce increased growth for
European Small and Medium-sized Enterprises (SMEs). X-
Scores provide actionable analytics in the shape of unique
metrics calculated out of the Natural Language Processing
(NLP) output. SSIX will extract meaningful financial signals
in a cross-lingual fashion from a multitude of social net-
work sources, such as Twitter, Google+, Facebook, StockTwits
and LinkedIn, and also authoritative news sources, such as
Newswires, Bloomberg, Financial Times and CNBC news
channel; transforming these signals into clearly quantifiable
sentiment metrics and indices regardless of language or lo-
cale. Financial services’ SMEs can customise SSIX indices
enabling them to provide meaningful domain specific insight
to design more efficient systems, test trading and investment
strategies, better understand risk and volatility behaviour of
social sentiment and identifying new investment opportunities.
Figure 1. SSIX platform architecture
SMEs can exploit the open source SSIX tools and method-
ologies to provide financial analytics services or alternatively
resell custom SSIX Indices as valuable financial data products
to third parties, thus leading to growth and increased revenue
for SSIX industry partners within the consortium and beyond.
Beyond the financial application, the SSIX approach and
methodologies can have broader impact across geopolitical and
socio-economic domains, generating multifaceted and multi-
domain sentiment index data for commercial exploitation. Fig-
12 Copyright (c) IARIA, 2016. ISBN: 978-1-61208-457-2
ALLDATA 2016: The Second International Conference on Big Data, Small Data, Linked Data and Open Data (includes KESA 2016)
22/129
ure 1 presents the overall SSIX platform architecture design.
The objectives of the SSIX project are to:
1) Develop the “X-Scores” statistical framework,
which will analyse metadata from indexed textual
sources to capture the signature of social sentiment,
generating a sentiment score. Statistical methods will
include regression, covariance and correlation analy-
sis. These X-scores will be used to create the custom
SSIX Indices.
2) Create an open-source template for generating
custom SSIX indices that can be tailor-made with
domain specific data parameters for specific analysis
objectives, such as Economics, Trading, Investing,
Government, Environmental or Risk profiling.
3) Create a powerful, easy to implement and low
latency “X-Scores API” to distribute the raw senti-
ment data feed and/or custom SSIX Indices that will
allow end users to easily integrate SSIXs sentiment
data into their own systems.
4) Enable end users to do cross-lingual target and
aspect oriented sentiment analysis over any sig-
nificant social network using user defined dedicated
SSIX Index.
5) Enable various public/private organisations and
institutions to create a SSIX Index and integrate
them with their proprietary tools in an easy to use
manner.
6) Explore the domain of SSIX Indices and X-Scores
beyond its primary focus of Finance applications.
Research has shown there is a positive correlation
between social media sentiment and a financial secu-
rities performance, but it is more difficult to measure
a broad topic such as, welfare of a region or commu-
nity. X-Scores will seek to provide metrics which can
filter out the noise and provide real quantifiable data,
which can give insight via a custom SSIX Index into
domains diverse as Education (SSIX-EDU), Media
trust (SSIX-MEDIA), Economic sociology (SSIX-
ECOSOC), Security (SSIX-SEC) and Health (SSIX-
HLTH).
7) Empower and equip SMEs within the emerging
Big Data Financial News sector to better com-
pete with established industry players via technology
transfer involving stable, mature and scalable open
source semantic and content analysis technologies.
8) Trigger, nurture and maintain a SSIX and X-
Scores commercial ecosystem within and beyond
the project lifecycle.
9) Pierce language barriers with respect to untapped
and siloed multilingual financial sentiment content
by harvesting cross lingual Big Social Media and
News Data.
By number crunching news text and social networks data
feeds regarding a company, product or various financial prod-
ucts (such as, stocks, funds, exchange-traded funds (ETFs),
bonds etc.) in a mathematical and statistical way, our approach
will allow investors and traders to combine SSIX generated
indices with their own proprietary tools and methodologies.
We envisage empowering the end-user, such as financial
data providers, financial, institutions, investment banks, wealth
management houses, asset management professionals, online
brokers, professional traders and individual investors with the
ability to make more informed and better and safer financial
decisions. Finally, SSIX could help in identifying unwanted or
dangerous trends that could be signalled to financial regulators
in advance in order to take appropriate measures, potentially
preventing unhealthy and toxic trading behaviour, thereby
safeguarding economic growth and prosperity.
The remainder of the paper discusses related work in
Section II. Information about SSIX Templates is provided in
Section III, whereas Section IV discusses Big Social and News
Data Management for SSIX. Details about Natural Language
Processing Services and Analysis is presented in Section V.
A Business Case Study about Investment and Trading is
discussed in Section VI, before providing some concluding
remarks in Section VII.
II. RELATED WORK
A. Sentiment Analysis on Financial Indices
In [2], Bormann defines several psychological definitions
about feelings, in order to explain what might be meant by
“market sentiment” in literature on sentiment indices. This
study is very relevant to SSIX, since it relates short and long
term sentiment indices to two distinct parts of sentiments,
namely emotion and mood; and extracts two factors repre-
senting investor emotion and mood across all markets in the
dataset.
The FIRST project [3] provides sentiment extraction and
analysis of market participants from social media networks in
near real-time. This is very valuable towards detecting and
predicting financial market events. This project is relevant to
SSIX, since the tool consists of a decision support model based
on Web sentiment as found within textual data extracted from
Twitter or blogs, for the financial domain. The relationship be-
tween sentiment and trading volume can provide the end-user
with important insights about financial market movements. It
can also detect financial market abuse, eg, price manipulation
of financial instruments from disinformation. Unlike SSIX,
only social networking services are used for extracting and
analysing sentiment, whereas the developed tool cannot be
easily customised to support media sources, target specific
companies or select the required language. In this respect,
SSIX provides a template methodology and source code to
create in a consistent manner the sentiment index for any type
of financial product and financial derivatives. Also the outcome
is easily integrated within other analytics tools as a data stream
with values between 0 and 100 that will define the ranges of
that specific sentiment.
Mirowski et al.[4] presents an algorithm for topic mod-
elling, text classification and retrieval from time-stamped doc-
uments. It is trained on each stage of its non-linear multi-
layer model in order to produce increasingly more compact
representations of bags-of-words at a document or paragraph
level, hence performing a semantic analysis. This algorithm
has been applied to predict the stock market volatility using
financial news from Bloomberg. The volatility considered is
estimated from daily stock prices of a particular company.
On a similar level, in [5] the authors present StockWatcher
through a customised, aggregated view of news categorised
by different topics. StockWatcher performs sentiment analysis
on a particular news messages. Each message can have either
a positive, negative or neutral effect on the company. This
13 Copyright (c) IARIA, 2016. ISBN: 978-1-61208-457-2
ALLDATA 2016: The Second International Conference on Big Data, Small Data, Linked Data and Open Data (includes KESA 2016)
23/129
tool enables the extraction of relevant news items from RSS
feeds concerning the NASDAQ-100 listed companies. The
sentiment of the news messages directly affects a company’s
respective stock price. SSIX, will extract meaningful financial
signals from multilingual heterogeneous (micro-blogging and
conventional) content sources and not just news items.
Gloor et al. introduces a novel set of social network
analysis based algorithms for mining unstructured information
from the Web to identify trends and the people launching
them [6]. This work is relevant, since the result of a three-
step process produces a “Web buzz index” for a specific
concept that allows for an outlook on how the popularity of
the concept might develop in the future. A possible application
of this system might be for financial regulators who try to
identify micro-and macro-trends in financial markets, eg,
showing the correlation between fluctuations in the Web buzz
index for stock titles and stock prices. Similarly, the Financial
Semantic Index estimates the probability that on a particular
day, an article in the financial press expresses a positive
attitude towards financial markets. This is measured through
the emotional tone of the mentioned article [7]. It is relevant
to SSIX, since it provides a certain viewpoint of the media
environment the market participants consume. In the case of
SSIX, it targets to transform the extracted information into
multiple clearly quantifiable social financial sentiment indices
regardless of language and data format. This will improve the
trading and investment accuracy through the combination of
various fundamental and technical parameters together with
sentiment ones.
B. Cross-lingual mining of information
The MONNET project provides a semantics-based solution
for integrated information access amongst language barriers
[8]. MONNET is relevant for SSIX, since one of its major
innovations is the provision of cross-lingual ontology-based
information extraction techniques for semantic-level extraction
of information for text and (semi) structured data across lan-
guages by using multilingual localised ontologies. It provides
real-life applications that demonstrate the exploitation potential
in several areas, such as financial services. In fact, one of the
project’s use-cases deals with searching and querying for finan-
cial information in the user’s language of choice. On the other
hand, it focused on cross-lingual domain, thus failed to target
other important aspects, eg, mining the extracted information.
SSIX will help identify unwanted/dangerous trends that could
be signalled to financial regulators in advance, in order to
potentially prevent unhealthy trading behaviour. Hence, SSIX
indices can be used as ‘early warning’signals for traders,
investors and regulator agencies, such as European Central
Bank, EU states national banks and rating agencies.
TrendMiner, another European project [9], presents an in-
novative and portable open-source real-time method for cross-
lingual mining and summarisation of large-scale social media
streams, such as weblogs, Twitter, Facebook, etc. One high
profile case study was a financial decision support (with ana-
lysts, traders, regulators and economists). In terms of novelty,
a weakly supervised machine learning algorithm is utilised for
automatic discovery of new trends and correlations, whereas
a cloud-based infrastructure is used for real-time text mining
from stream media. This project is relevant to SSIX given
that it provides several multilingual ontology-basedsentiment
extraction methods.
The main goal of the LIDER project [10] is to create a
Linguistic Linked Data (LLD) cloud that is able to support con-
tent analytics tasks of unstructured multilingual cross-media
content. This will help in providing an ecosystem for a new
Linked Open Data based ecosystem of free, interlinked and
semantically interoperable language resources (eg, corpora,
dictionaries, etc.) and media resources (eg, image, video,
etc.). It also aims to make an impact on the ease and efficiency
with which LLD is exploited in processes related to content
analysis with several use cases in multiple industries within the
areas of social media, financial services and other multimedia
content providers and consumers. One limitation is that LIDER
aims to make an impact on the LOD cloud and not to further
transform any extracted signals into clearly quantifiable social
sentiment indices, as in the case of SSIX. Such indices are
targeted to any equities, stock indices or derivatives.
The AnnoMarket project has delivered a cloud-based plat-
form for unstructured data analytics services, in multiple
languages [11]. This text annotation market is delivered via
annomarket. com and has been in public beta as of April 2014.
The services being offered can be adopted and applied for
many business applications, eg, large-volume multi-lingual
information management, business intelligence, social media
monitoring, customer relations management. It includes several
text analytics services that would be of benefit to the SSIX
project. Similarly, OpeNER will provide a number of ready
to use tools in order to perform some NLP tasks (entity
mentions detection and disambiguation, sentiment analysis and
opinion detection) that can be freely and easily integrated
in the workflow of SMEs [12]. this project aims to have
a semi-automatic generation of generic multilingual (initially
for the English, French, German, Dutch, Italian and Spanish
languages) sentiment lexicons with cultural normalisation and
scales through the reuse of existing language resources. SSIX
goes beyond text analysis on unstructured data, since an “X-
Scores” statistical framework will be implemented to capture
the signature of social sentiment from indexed textual sources.
These scores will help create custom SSIX Indices that can
be tailored for a particular domain depending on specific data
parameters. This will provide a meaningful insight to drive
trading, investment decisions and strategies, and create new
investment opportunities.
III. SSIX TEMPLATES
SSIX templates will empower both the public and private
sectors to develop innovative disruption-enabling mobile and
cloud services and products, to leverage the massive amount
of sentiment data that is constantly produced and published on
various social media networks within multiple domains such
as Finance, Economy, Government, Politics and Health.
The SSIX templates will be able to gauge the actual voiced
sentiment from social media conversations, specifically emo-
tional attributes, such as (but not restricted to) optimism and
pessimism. These sentiment signals can be analysed to evaluate
their influence on real world financial/economic/social/political
outcomes and can act as valid indicators. An ideal paradigm
that can benefit from the integration of SSIX templates is
the field of investment decisions. Traditionally, research on
securities, such as stocks, fixed income and foreign exchange
14 Copyright (c) IARIA, 2016. ISBN: 978-1-61208-457-2
ALLDATA 2016: The Second International Conference on Big Data, Small Data, Linked Data and Open Data (includes KESA 2016)
24/129
relied on applying a Fundamental and/or Technical Analysis
approach to determine the most efficient and lowest risk
investment decision for a given amount of expected return. In
this scenario, market sentiment is derived from the aggregation
of a variety of these two disciplines (Fundamental and Tech-
nical analysis), including attributes, such as price action, price
history, economic and financial reports/data, market valuation
indicators, fund flows, sentiment surveys (eg, ZEW Indicator
of Economic Sentiment-A Leading Indicator for the German
Economy), commitment of traders report analysis, analysis of
open interest from the futures market, seasonal factors and
national/world events. As a consequence, it is difficult to get a
reliable and easy to interpret measure of a securities sentiment
score without using a selection bias and almost impossible
to measure a niche sector efficiently; this type of sentiment
classification tends to be a lagging indicator to price movement
but can act as confirmation.
The growth of social media APIs and the application of
news analytics has provided a new method allowing sentiment
analysis from a social media perspective to be carried out
on financial securities, which has been proven to show a
positive correlation to price performance (“Twitter is now
a leading indicator of movement (up and down) of specific
stocks-we can prove it.”, Social Market Analytics). This
data can be analysed to gain a greater understanding of
sentiment behaviour and its correlation to price volatility for
an individual security/sector or the entire market. By using this
new sentiment data source, SSIX can deliver unique sentiment
indices using X-Scores (a statistical framework of qualitative
and quantitative parameters, such as regression, covariance
and correlation analysis), such as the ‘Social Sentiment Index
for Healthcare’-SSIX Health or the ‘Social Sentiment Index
for Technology’-SSIX Tech, which will show the sentiment
levels for their corresponding sectors, quantifying how market
participants feel. X-Scores metrics can used in conjunction
with industry standard technical parameters to analyse se-
curities, such as Moving Average Convergence-Divergence
(MACD), Relative Strength Index (RSI), Moving Averages
(MA), Exponential Moving Average (MVA), Pivots Points, etc.
SSIX X-Scores will provide real quantifiable data and tools to
anticipate volatility and to analyse past performance, which
will help develop alternative and more efficient approaches
to reduce risk. SSIX can be used to identify trading signals,
helping to make more informed investment decisions, resulting
in a more efficient use of capital while reducing any associated
risk. SMEs will be able to integrate the SSIX framework data
into their own models for use in any area of application where
sentiment analysis is used.
IV. BIG SOCIAL AND NEWS DATA MANAGEMENT
Data retrieved from digital social networking and news
sources provides significant data samples to the NLP com-
ponent of SSIX. The entire process is developed through the
following steps:
• Data download and gathering from different digital
platforms (social networks, blogs, news sites, etc.)
with different techniques (API usage, CSV download,
Web scraping, etc.);
• Data cleaning and filtering to isolate significant infor-
mation;
• Data processing to produce analysed and enriched data
(smart data);
• Data sampling to extract pieces of smart data intended
to be used by NLP component.
A. Big Data Challenges
In SSIX, multiple kinds of data are constantly collected,
which process is continuous for the duration of the project.
The following are types of data in question:
• Public available data from social networks
• Datasets part of the Linking Open Data (LOD) cloud
• LLD Cloud resources
• Public data available from domain-specific SMEs
• Survey data collected from independent events, such
as technology summits, conferences, etc., or organised
events, such as workshops, focus groups, etc.
• Financial and Economic trends outlined by the SSIX
framework from analysis/mining of data
• Language Resources (LRs) either automatically ac-
quired or reused from SentiWordNet (LR for opinion
mining) and EuroSentiment (EU Project that provides
a marketplace for LRs and Services dedicated to
Sentiment Analysis).
Several challenges also arise due to the diverse nature
of the gathered data. SSIX is able to deal with the three
main challenges coming from the big data field namely, high
volume, high velocity and high variety.
• High volume: constant growing of the data repository
is managed through adoption of scalable technologies
and architectures. The space required for the storage
can be easily increased on request, while the technolo-
gies used are suitable to manage big quantities of data
(eg, Cassandra, Hadoop).
• High velocity: big stream of data is collected and man-
aged with specific technologies and adequate process-
ing capabilities. The project adopts high-performing
servers with possibility to scale the computing power.
• High variety: the gathered data comes from multiple
sources. In this case, each data source is treated
separately. When required, an unstructured data model
is implemented, in order to store information that can
vary over time.
B. From Big Data to Smart Data
Figure 2 illustrates the flow that all the data will follow be-
fore entering the SSIX platform for further NLP and analysis,
which process transforms the data retrieved into smart data.
Each process is explained in more detail as follows:
• BIG DATA: indicates all the information available on
different external platform in form of data sources
(eg, social networks, blogs, news sites, etc.)
• DOWNLOADER: the data are gathered from the
different data sources using techniques, such as API
usage, CSV download and parsing, web pages scrap-
ing, etc.
15 Copyright (c) IARIA, 2016. ISBN: 978-1-61208-457-2
ALLDATA 2016: The Second International Conference on Big Data, Small Data, Linked Data and Open Data (includes KESA 2016)
25/129
• DATA FILTERING→ FILTERED DATA: a first pro-
cess of noise removal and data processing that will
produce a layer of filtered data.
• DATA PROCESSING→ SMART DATA: in this phase
of the process, all the data will be parsed and trans-
formed into smart data.
• DATA SAMPLING→ SAMPLES FOR NATURAL
LANGUAGE PROCESSING: the last step will consist
in the extraction of significant data samples destined
for NLP.
Figure 2. SSIX platform data-flow
All the smart data will be archived into a high performing
repository. A cluster of servers will produce significant samples
retrieved from the smart data repository that will be taken and
streamed to the SSIX platform by and End Point component.
The first prototype will use three physical servers to implement
the architecture presented in Figure 3.
Figure 3. SSIX platform data architecture
The schema defined in Figure 3 illustrates the ideal ar-
chitecture delegated to retrieve data from the identified data
sources, in order to process it and to create data samples for
the NLP phase. The business case studies that will be executed
in the duration of the project (such as the one discussed in
Section VI) will be managed by a cluster of machines that will
include: i) a software component that will interface with the
different data sources, which will retrieve the data from them;
ii) a repository of filtered data; and iii) a software component
for data processing.
V. NATURAL LANGUAGE PROCESSING SERVICES AND
ANALYSIS
Analysing trends in social media content results in the
process of a very large number of comparably short texts
in near real-time. Therefore, the major challenge for the
implementation of the NLP pipeline is in the orchestration of
the different analysis components in a way that is potentially
scalable in a cluster of servers that is able to handle hundreds
of messages per second. Special care has to be taken to provide
the NLP process as a distributed near real-time computation
system that can reliably process unbounded streams of data.
SSIX implements this process based on Apache Storm. Apache
Storm is a framework that offers the foundations of distributed
stream processing and is also fault-tolerant. Moreover, SSIX
addresses the following major objectives:
• Automatic execution planning of NLP analysis pro-
cesses: based on the descriptions of existing analysis
components, available input and infrastructure, and de-
sired output, SSIX automatically computes an appro-
priate execution plan (“topology” in Apache Storm);
• Standardised API for analysis components: a common
problem in NLP processing is that there are many
components for different, but related tasks, but they
all implement completely different APIs, making it
hard to combine them efficiently in a process. SSIX
provides a standardised API and a standardised com-
ponent description format to simplify integration of
existing and additional analysis components.
• Sufficient collection of initial components: a big chal-
lenge in building this pipeline is to provide a sufficient
collection of initial components so that we can (1)
validate our execution model and API, and also pro-
vide examples for developers,(2) provide a process
for real-time analytics, and (3) integrate with queuing
and database technologies provided by SSIX. Figure 4
provides an overview architecture of the NLP pipeline.
A. Multilingual Language Resource Acquisition and Manage-
ment
The multilingual language resource acquisition and man-
agement occurs in two phases:
1) Identification and resource of existing language re-
sources for adaption for SSIX business cases (one
business case example is discussed in more detail
in Section VI), ie exploitation of multilingual sen-
timent and domain specific lexica from European
projects, such as EuroSentiment [13]–which pro-
vides a shared language resource pool for fostering
sentiment analysis from a multilingual, quality and
domain coverage perspective–or the adaptation of
16 Copyright (c) IARIA, 2016. ISBN: 978-1-61208-457-2
ALLDATA 2016: The Second International Conference on Big Data, Small Data, Linked Data and Open Data (includes KESA 2016)
26/129
LLD resources and carry out any necessary localisa-
tion of monolingual resources where target language
equivalents are scarce, such as Asian languages.
2) Exploration of unsupervised and/or semi-automatic
corpus based methods for acquisition of multilingual
lexica to support entity and sentiment analysis tasks.
Figure 4. SSIX platform Knowledge-based NLP pipeline
VI. BUSINESS CASE STUDY: INVESTMENT AND TRADING
The SSIX sentiment index template will be used to deter-
mine social sentiments on stocks and then incorporate them as
independent parameters within Peracton’s MAARS platform
(www. peracton. com/maars), for complex evaluation together
with other financial parameters. Performance analysis will be
made over historic data and real time, to determine the impact
of SSIX indices. This case study will be made of the following
5 phases:
Phase 1: Establish data sources and targets in order to
generate unique SSIX Indices
Establish data sources and targets in order to generate
unique SSIX Indices In this phase we will identify the
suitable data sources available on various social networks
(Twitter, Facebook, LinkedIn, Google+). It is estimated that
approximately 6000+ stocks will be traced individually on
US exchanges, such as NASDAQ, AMEX and NYSE.
Phase 2: SSIX indices generation and storage
Once the data sources are established, SSIX engine will
be instantiated to generate 6000+ unique sentiment indices
that trace 6000+ US stocks. Such indices will be uniquely
identified, such as SSIX AAPL, SSIX YHOO, SSIX LNKD,
SSIX FB, etc. Once instantiated, the SSIX indices values will
be generated first for every day and stored accordingly and
then for every minute (if this will be technically feasible).
Phase 3: SSIX indices integration within MAARS
The 6000+ generated index sentiment values, stored every
day (and every minute) will be integrated within MAARS
analytics and attached to the existing financial data stocks that
are already stored within MAARS cloud.
Phase 4: Trading and Investment with SSIX indices
As sentiment data starts to be updated within MAARS
analytics, simulations tests of investing and trading will be
performed. There will be trading and investment tests with no
SSIX sentiment data (control tests) and then in parallel, same
investment and trading tests involving sentiment data.
Phase 5: Feedback Based upon Phase 4 tests, feedback
will be provided to the performance and changes in results of
investment/trading exercise due to using sentiment data.
VII. CONCLUSION
SSIX seeks to extract and measure meaningful financial
sentiment signals in a cross-lingual fashion, from a vast
multitude of social network sources, such as Twitter, Face-
book, StockTwits, LinkedIn and public media outlets, such
as Bloomberg, Financial Times and CNBC. It will generate
custom X-Scores powered index for a given sentiment target
or aspect, ie company or financial product. The primary
domain is finance although SSIX has scope for Environment,
Health, Technology, Geopolitics and beyond. The X-scores
will be used by the industrial partners and bundled with
their financial analytics, in order to increase the accuracy of
their output combined with either end of day financial data
or, real time data feeds. SSIX will adapt existing mature,
proven and scalable open source text mining tools in order
and circumvent language barriers with respect to unexploited
multilingual financial sentiment content by harvesting cross
lingual Big Social Media and News Data. Semantic Analytics
will be employed to generate SSIX indices.
ACKNOWLEDGMENT
This project has received funding from the European
Union’s Horizon 2020 Research and Innovation Programme
ICT 2014-Information and Communications Technologies
under grant agreement No. 645425.
REFERENCES
[1] D. Greenfield,“Social media in financial markets: The coming of age...”
GNIP, GNIP Whitepaper, 2014.[Online]. Available: http://stocktwits.
com/research/social-media-and-markets-the-coming-of-age. pdf
[2] S.-K. Bormann,“Sentiment indices on financial markets: What do they
measure?” Kiel Institute for the World Economy, Economics Discussion
Paper 2013-58, 2013.[Online]. Available: http://www. economics-
ejournal. org/economics/discussionpapers/2013-58
[3]“FIRST-large scale inFormation extraction and Integration infrastruc-
ture for SupporTing financial decision-making,” http://project-first. eu/,
2013.
[4] P. Mirowski, M. Ranzato, and Y. LeCun,“Dynamic auto-encoders
for semantic indexing,” in NIPS 2010 Workshop on Deep Learning,
Proceedings.
[5] A. Micu, L. Mast, V. Milea, F. Frasincar, and U. Kaymak,“Financial
news analysis using a semantic web approach,” in Semantic Knowledge
Management: an Ontology-based Framework, Paolo Ceravolo, Ernesto
Damiani, Gianluca Elia, Antonio Zilli (Eds.), November 2008, pp. 311–
328.
[6] PA Gloor, J. Krauss, S. Nann, K. Fischbach, and D. Schoder,
“Web Science 2.0: Identifying Trends through Semantic Social Network
Analysis,” in International Conference on Computational Science and
Engineering. CSE’09., pp. 215–222.
[7] Ontology2,“FSI: Financial Semantic Index,”
http://financialsentimentindex. com/fsi/, 2012.
[8]“MONNET-Multilingual Ontologies for Networked
Knowledge,” http://cordis. europa. eu/fp7/ict/language-
technologies/projectmonnet en. html, 2013.
[9]“TrendMiner,” http://www. trendminer-project. eu/, 2014.
[10]“LIDER:“Linked Data as an enabler of cross-media and
multilingual content analytics for enterprises across Europe”,”
http://www. liderproject. eu/, 2015.
[11]“AnnoMarket,” https://annomarket. eu/, 2014.
[12]“OpeNER,” http://www. opener-project. org/, 2014.
[13]“EuroSentiment,” http://eurosentiment. eu/, 2014.
17 Copyright (c) IARIA, 2016. ISBN: 978-1-61208-457-2
ALLDATA 2016: The Second International Conference on Big Data, Small Data, Linked Data and Open Data (includes KESA 2016)