Content uploaded by Barbara Calabrese
Author content
All content in this area was uploaded by Barbara Calabrese on Dec 18, 2016
Content may be subject to copyright.
Using social networks data for behavior and
sentiment analysis
Mario Cannataro?, Nicola Ielpo, and Barbara Calabrese
University ”Magna Graecia”
viale Europa, 88100, Catanzaro, Italy
Abstract. With the advent of social networks every day, it is generated
and stored a huge amount of information. The social network therefore
represents a potentially infinite source of user data, usable both for sci-
entific and commercial applications. Specifically, they store a lot of data
regarding individual and collective behavior, and informations relative
to individuals and their relationship with other users. The combination
of behavior and sentiment analysis tools with methodologies of affective
computing could allow the extraction of useful data that convey infor-
mation for different applications, such as depression state detection in
psychology field.
Keywords: social network, behavior analysis, sentiment analysis, affec-
tive computing
1 Social Networks: a main data source for detecting
behaviors and emotions
In recent years, social networks have seen an exponential increase and now they
store a lot of data regarding individuals and their relationship with other indi-
viduals as well as data regarding groups or communities of users. Thus, analyzing
social network data may be helpful to detect behaviour and emotions of individ-
ualsa as well as well of communities.
Social networks are the most representative models of Graph Databases. In
social networks [26], the nodes are people and groups, while the links show
relationships or flows between the nodes. Some examples are the friendships,
business relationships, research networks (collaboration, co-authorship), record-
ings of communications (mail, phone calls, etc . . . ), computer networks, national
security. There is an increasing activity in the field of social networks analysis,
visualization and data processing in such networks.
Social networks provide some interfaces that allow to access to the data of its
members in respect of their privacy. However, since they base their own business
model on these data, consequently the access modes (legal) to the data and the
amount of data extracted from the social network are very low. Alternatively,
?Corresponding author
2 Mario Cannataro et. al.
there are other methods for extracting data from social network associated with
the Web Scraping.
To analyze social networks data the following main tasks need to be per-
formed: social network data extractions (e.g. through the APIs provided by ma-
jor social networks providers); data storage (e.g. through the emerging NoSQL
databases well suited to store graph-based data); data analytics (to discover be-
haviors and emotions) eventually coupled with complementary approaches such
as affective computing, eye movement and facial movement detection.
Below are shown the main features of the two most popular social network:
Facebook and Twitter.
1.1 Facebook API
Facebook is a social network that allows registered users to create their own
profile to upload photos, videos and send messages, in order to keep in touch
with friends, family and colleagues. It also has the following features: groups,
events, pages, chat, marketplace and privacy manager. In order to try to extract
information from a user, you must be registered to the social network. If the user
belongs to my friends network then I can have access to all the information that
he has made available on its user profile. Instead in the case in which he is not
my friend, the data that I might be able to collect depend entirely on how he
has set the privacy for his profile. So with regard to Facebook, find information
about a user that is not in our friends network turns out to be a very complicated
operation and appears to be influenced by the level of privacy that this user has
set for herself. Generally most of the registered users on Facebook do not change
these settings and leave the default ones that allow the tracing of the person on
the search engines and leave public the basic information of the profile.
Facebook is characterized by the social graph, the graph in which nodes rep-
resent entities (such as people, pages or applications) and arcs represent entities
connections. Therefore each entity or object is a node in the social graph and
every action of an entity is an arc. It is possible interact with the social graph in
a unique way, ie through HTTP calls to the Facebook API [21]. The interaction
involves two main components: the Graph API for reading and writing of the
social graph and the OGP (Open Graph Protocol), which is the protocol that
allows you to insert any object in the social graph by simply entering within it
the meta-data RDFa (Resource Description Framework in Attributes).
The Graph API allow you to have a uniform view of the Facebook social
graph through simple HTTP calls; it is obtained, therefore, a subset of the
nodes and connections of the graph. Each entity of the social graph has a
unique identifier; then the properties of the object can be tracked at https :
//graph.f acebook.com/ < id >. Alternatively, all entities with a username field
(such as user profiles and pages) are accessible using this field instead of the id.
HTTP calls return a response message with a well-defined structure (XML or
JSON); the latter is the most widely used format for response messages of Web
services because it is much lighter than XML. It is possible to examine the con-
nections between the various objects of the social graph using the following URL
Using social networks data for behavior and sentiment analysis 3
structure:https ://graph.f acebook.com/ < id > /connection type. The request
URL may include the access token parameter in case in which it is expected the
authenticated and authorized access process of to the information contained.
The Graph API therefore allow to easily access all public information of
an object; in case you wish to obtain additional information, you need permis-
sion of the entity that owns them. As previously described, the authentication
mechanism of Facebook is based on OAuth 2.0 protocol, which provides for the
acquisition of an access token. There are different ways of acquiring the access
token; the easiest is to go into the Graph API Explorer and press the ”Get
Access Token”. Then you will need to select the permissions you are interested
in selecting the appropriate boxes. Of course this can also be done via HTTP,
just hanging to the query string of the API HTTP address the access token
parameter (https ://graph.f acebook.com/me?access token =...).
The Graph API provide three types of permits: base permits (do not require
any access token); user data and friends data permits (they are designed to
restrict access to personal data of users); extended permits (required for publi-
cation and access to sensitive data). So, the access token is a mechanism whose
aim is to provide a temporary and limited access to the Facebook API. Graph
API calls return most of the properties of the object of the social graph related
to the query sent; to select the parameters that you want returned you must
enter the fields parameter in the search string of the API call.
1.2 Twitter API
Twitter is a service of micro-blogging with two main characteristics: its users
send messages (tweets) of 140 characters usually compounds by keywords (in
the form of hashtags), natural language and common abbreviations; moreover,
each user can follow other users so that his timeline is populated by their tweets.
It is much easier to obtain user data because the profiles are public and can be
viewed by anyone. As for Facebook there is the ability to change the settings
relating to privacy so that users can see their profile only after we have accepted
their request. Even in this case, however, users who choose this route about pri-
vacy are few; moreover also any person not registered in Twitter can access to
user profiles. Both Social Network provide APIs (Application Programming In-
terface) with OAuth authentication mechanism. The purpose of this protocol is
to provide a framework for the verification of the identity of the entities involved
in secure transactions. There are at the moment two versions of this protocol:
OAuth 1.0 [19] and OAuth 2.0 [20]. Both versions support two-legged authenti-
cation, in which a server is guaranteed about the user identity and three-legged
authentication, in which a server is guaranteed by an application about the user
identity. This type of authentication requires the use of the access token and it
is currently implemented by the Social Network.
Compared to Facebook, Twitter connections are bidirectional: there is an
asymmetric network consisting of friends, that is the accounts that a user follow
and followers, that is the accounts that follow the user. The timeline of a user
4 Mario Cannataro et. al.
that you can trace in the home consists of a real-time stream containing all the
tweets of his friends.
As well as Facebook provides the Graph API Explorer useful to explore the
API, Twitter provides the Twitter Console; generally Twitter offers an extensive
collection of APIs, all based on HTTP [22]. Twitter supports two authentication
methods based on the OAuth protocol: the first one based on OAuth 1.0a related
to the user and the second one based on OAuth 2.0 related to an application.
The first mode defined application-user authentication includes an HTTP
authorization request that communicates what application is making the request,
on behalf of which user the application is making the request, if the user has
authorized or not the application and if during transit the request has been
tampered by third parties.
In the second mode, defined application-only authentication, the application
encodes its consumer key and its secret key in a set of encoded credentials and
then performs an HTTP POST request to endpoint oauth2/token to exchange
these credentials with a bearer token. The bearer token obtained is used to
authenticate the application that it represents in the REST API. The latter
approach is much simpler because it is not required that the call is signed.
The typology of the Twitter API end-point is the following: https ://api.twitter.com/1.1/ <
resource > / < action >. The Twitter API include 16 resources: timeline, tweet,
search, streaming, direct messages, friends and followers, users, user suggested,
favorites, lists, saved search, places, trends, spam reporting, OAuth, help. The
Twitter Search API allows the execution of real-time queries on recent tweet. In
particular, the query must be simple, limited to a maximum of 1000 characters,
including operators and it is always required some form of authentication. In this
case the only available resource is the tweet. The Twitter Streaming API allow a
real-time update of information relating to specific resource, thereby eliminating
the need to repeatedly call at regular intervals (polling) its REST end-point.
1.3 NoSQL Database
In recent years it has increased more and more the volume of data to be stored
and the need to process large amounts of data in a short time; therefore you
are heading for a new model of data management that moves away from the
relational model: the NoSQL model. It provides four main families of database:
Key-Values stores; Column-oriented databases; Document databases and Graph
databases [1]. Graph databases have become a topic of interest in the last years,
mainly due to the large amount of data modeled as a graph introduced by
web. A graph, the key element of the Graph database, is defined as a simple
mathematical structure that consists of nodes and arcs connecting the nodes.
More formally, a graph is an ordered pair of sets G = (V, E), with V a set of
nodes and E a set of arcs, such that the elements of E are elements of V pairs
[2].
Using social networks data for behavior and sentiment analysis 5
2 Behaviour Analysis
With the rise of social media, users are given opportunities to exhibit different
behaviors such as sharing, posting, liking, commenting, and befriending conve-
niently. By analyzing behaviors observed on social media, it is possible to classify
these behaviors into individual and collective behavior. Individual behavior is ex-
hibited by a single user, whereas collective behavior is observed when a group
of users behave together [18].
In [3], the authors investigated whether posts on FB would also be applica-
ble for the prediction of users psychological traits such as self-monitoring (SM)
skill that is supposed to be linked with users expression behavior in the online
environment. They present a model to evaluate the relationship between the
posts and SM skills. They first, evaluate the quality of responses to the Snyders
Self-Monitoring Questionnaire (inerire rif) collected via the Internet; and sec-
ondly, explore the textual features of the posts in different SM-level groups. The
prediction of posts resulted in an approximate 60% accuracy compared with the
classification made by Snyders SM scale. They concluded that the textual posts
on the FB Wall could partially predict the users SM skills.
Zhang et al. propose a socioscope model for social-network and human-
behavior analysis based on mobile-phone call-detail records. They use multiple
probability and statistical methods for quantifying social groups, relationships,
and communication patterns and for detecting human-behavior changes. They
propose a new index to measure the level of reciprocity between users and their
communication partners. For the validation of our results, we used real-life call
logs of 81 users which contain approximately 500 000 h of data on users’ loca-
tion, communication, and device-usage behavior collected over eight months at
the Massachusetts Institute of Technology (MIT) by the Reality Mining Project
group [17].
3 Sentiment Analysis
Sentiment analysis aims to analyze people’s sentiment, opinions, attitudes, emo-
tions. Different techniques and software tools have been developed to carry out
Sentiment Analysis. Most of works in this research area focus on classifying texts
according to their sentiment polarity, which can be positive, negative or neutral.
Therefore, it can be considered a text classification problem, since its goal con-
sists of categorizing texts within classes by means of algorithmic methods. The
paper [10] offers a comprehensive review about this topic and compares some
free access web services, analyzing their capabilities to classify and score different
pieces of text with respect to the sentiments contained therein.
In the last years, thanks to the increasing amount of information delivered
through social networks, many researches have been focused on applying senti-
ment analysis to these data [12], [13]. Sentiment analysis aims at mining users
opinion and sentiment polarity from the posted text on the social network.
In [7], the authors apply data mining techniques to psychology, specifically to
the field of depression, to detect depressed users in social network services. The
6 Mario Cannataro et. al.
create an accurate model based on sentiment analysis. In fact, the main symptom
of the depression is severe negative emotions and lack of positive emotions.
In [11], a new method for sentiment analysis in Facebook has been presented
aiming (i) to extract information about the users sentiment polarity (positive,
neutral or negative), as transmitted in the messages they write; and (ii) to model
the users usual sentiment polarity and to detect significant emotional changes.
The authors have implemented this method in SentBuk, a Facebook applica-
tion. SentBuk retrieves messages written by users in Facebook and classifies
them according to their polarity, showing the results to the users through an
interactive interface. It also supports emotional change detection, friends emo-
tion finding, user classification according to their messages, and statistics, among
others. The classification method implemented in SentBuk follows a hybrid ap-
proach: it combines lexical-based and machine-learning techniques. The results
obtained through this approach show that it is feasible to perform sentiment
analysis in Facebook with high accuracy (83.27%).
4 Affective Computing
Affective Computing is computing that relates to, arises from, or deliberately
influences emotion or other affective phenomena [9]. Existing emotion recogni-
tion technologies include physiological signals recording, facial expression and/or
voice analysis [14], [15]. Physiological emotion recognition is based on obtrusive
technologies that require special equipment or devices, e.g. skin conductance
sensors, blood pressure monitors, ECG and/or EEG recording devices. Facial
expressions and voice systems for emotion recognition, instead, use devices that
should be positioned in front of the face of the user or should always listen to the
voice of user [8]. In the following, some examples of emotion recognition systems
are reported and discussed.
C. Peter et al. proposed wearable system architecture for collecting emotion-
related physiological signals such as heart rate, skin conductivity, and a skin
temperature of users [4]. They developed a prototype system, consisting of a
glove with a sensor unit, and a base unit for receiving the data transmitted from
the sensor unit. S. V. Ioannou et al. realize an emotion recognition system based
on the evaluation of facial expressions [5]. They implemented a neuro-fuzzy net-
work based on rules which have been defined via analysis of facial animation
parameters (FAPs) variations of users. With experimental real data, they also
showed acceptable recognition accuracy of higher than 70%. A. Batliner et al.
presented an overview of the state of the art in automatic recognition of emo-
tional states using acoustic and linguistic parameters [6]. They summarized core
technologies such as corpus engineering, feature extraction, and classification
have been used for building emotion recognition systems via the speech analysis.
In [8], the authors present a machine learning approach to recognize emo-
tional states through the acquisition of some features related to user behavioral
patterns (e.g. typing speed) and the user context (e.g. location) in the social net-
work services. They developed an Android application that acquires an analyzes
Using social networks data for behavior and sentiment analysis 7
these features whenever the user sends a text message to Twitter. They built a
Bayesian classifier that recognizes seven classes: one neutral and six relative to
basic emotions with an accuracy of 67,52%.
The paper [16] describes an intelligent and affective tutoring system designed
and implemented within a social network. The tutoring system evaluates cogni-
tive and affective aspects and applies fuzzy logic to calculate the exercises that
are presented to the student. The authors use Kohonen neural networks to rec-
ognize emotions through faces and voices and multi-attribute utility theory to
encourage positive affective states.
5 Towards an integration of existing approaches
Social networks user often expresses her/his feeling or emotional states eith writ-
ten text or by using emoticon. However, some users have difficulties to express
their feelings or can simulate. These limits could overcome by adopting emotion
recognition technologies related to affective computing and combinining them
with typical text-mining methodologies of behavior and sentiment analysis.
References
1. Eifrem, E.: A nosql overview and the benefits of graph databases. Nosql East 2009
2. Trudeau, R. J.: Introduction to Graph Theory, Dover Publications, 1994
3. He, Q., Glas, C. A. W., Kosinski, M., Stillwell, D. J., Veldkamp, B. P.,: Predict-
ing self-monitoring skills using textual posts on Facebook. Computers in Human
Behavior. 33, 69–78 (2014)
4. Peter, C., Ebert, E., Beikirch, H.,: A wearable multi-sensor system for mobile ac-
quisition of emotion-related physiological data. Affective Computing and Intelligent
Interaction. 3784, 691-698 (2005)
5. Ioannou, S. V., Raouzaiou, A. T., Tzouvaras, V. A., Mailis, T. P., Karpouzis, K.
C., Kollias,S. D.,: Emotion recognition through facial expression analysis based on
a neurofuzzy network. Neural Networks. 18:4, 423-435, (2005)
6. Batliner, A.,Schuller, B., Seppi, D., Steidl, S., Devillers, L., Vidrascu, L., Vogt,
T., Aharonson, V., Amir, N.,: The automatic recognition of emotions in speech.
Emotion-Oriented Systems. 2, 71-99, (2011)
7. Wang, X., Zhang, C., Ji, Y., Sun, L., Wu, L.,: A Depression Detection Model Based
on Sentiment Analysis in Micro-blog Social Network. Trends and Applications in
Knowledge Discovery and Data Mining. LNCS, vol. 7867, 201–213, (2013)
8. Lee, H., Choi, Y. S., Lee, S., Park, I. P.,:Towards Unobtrusive Emotion Recognition
for Affective Social Communication. In: 9th Annual IEEE Consumer Communica-
tions and Networking Conference, pp. 260–264. IEEE, (2012)
9. Picard, R.,: Affective Computing. Cambridge MIT Press, (2000)
10. Serrano-Guerrero, J., Olivas, J. A., Romero, F. P., Herrera-Viedma, E.,: Sentiment
analysis: A review and comparative analysis of web services. Information Sciences.
311, 18-38, (2015)
11. Ortigosa, A., Martin, J. M., Carro, R. M.,: Sentiment analysis in Facebook and its
application to e-learning. Computers in Human Behavior. 31, 527-541, (2014).
8 Mario Cannataro et. al.
12. Go, A., Bhayani, R., Huang, L.,: Twitter sentiment classification using distant
supervision. Technical report. Stanford University. Stanford Digital Library Tech-
nologies Project, (2009)
13. Pak, A., Paroubek, P.,: Twitter as a corpus for sentiment analysis and opinion
mining. In Proceedings of the seventh conference on international language resources
and evaluation, pp. 1320-1326, (2010)
14. Armony, J. L.,: Affective Computing. Trends in Cognitive Sciences. 2(7), 270,
(1998)
15. Calvo, R.A., D’Mello, S.,: Affect Detection: An Interdisciplinary Review of Models,
Methods, and Their Applications. Affective Computing, IEEE Transactions on. 1(1),
18–37, (2010)
16. Barrn-Estrada, M. L., et al.: An Intelligent and Affective Tutoring System within
a Social Network for Learning Mathematics. Advances in Artificial Intelligence IB-
ERAMIA 2012. LNCS Volume 7637, pp. 651–661 (2012)
17. Zhang, H., Dantu, R., Cangussu, J.W.,: Socioscope: Human Relationship and Be-
havior Analysis in Social Networks.Systems, Man and Cybernetics, Part A: Systems
and Humans, IEEE Transactions on. 41(6), 1122–1143, (2011)
18. Zafarani R., Liu, H.,: Behavior Analysis in Social Media. IEEE Intelligent Systems.
29(4), (2014)
19. Hammer-Lahav, E.,: The OAuth 1.0 Protocol. RFC 5849, April 2010
20. Hardt, D.,: The OAuth 2.0 Authorization Framework. RFC 6749, October 2012
21. Facebook. Graph API - Facebook Developers.
https://developers.facebook.com/docs/reference/api/, Mar. 2012
22. Twitter API. https://dev.twitter.com/overview/documentation
23. Catanese, S.A., De Meo P., Ferrara, E.,: Crawling facebook for social network anal-
ysis purposes. In: Proceedings of the International Conference on Web Intelligence,
Mining and Semantics (2011)
24. Traud, A. L., Kelsic, E. D., Mucha P. J., Porter, M. A.,: Comparing Community
Structure to Characteristics in Online Collegiate Social Networks. SIAM review
53(3): 17 (2008)
25. Traud, A. L., Mucha P. J., Porter, M. A.: Social Structure of Facebook Net-
works.CoRR: 82 (2011)
26. Hanneman, R. A.,: Introduction to Social Network Methods. Technical report,
Department of Sociology, University of California, Riverside (2001).
