Coletto, Mauro (2017) Analysis of Polarized Communities in Online Social Networks. Advisor: Lucchese, Dr. Claudio. Coadvisor: De Nicola, Prof. Rocco . pp. 207. [IMT PhD Thesis]
|
Text
Coletto_phdthesis.pdf - Published Version Available under License Creative Commons Attribution No Derivatives. Download (35MB) | Preview |
Abstract
Increasingly, people around the globe use Social Media (SM) - e.g. Facebook, Twitter, Tumblr, Flickr, Youtube - to publish multimedia content (posting), to share it (retweeting, reblogging or resharing), to reinforce it or not (liking, disliking, favoriting) and to discuss (through messages and comments) in order to be in contact with other users and to get informed about topics of interest. The world population is ≈ 7:4 billion people, among them ≈ 2:3 billion (31%) are active social media users (GlobalWeb Index data, Jan 2016). In fact, these virtual contexts answer the human need of aggregation that nowadays is translated into digital bonds among peers all over the world, in addition to the traditional face-to-face relationships. Online Social Networks (OSNs), then, provide a space for user aggregation in groups, expressing opinions, accessing information, contributing to public debates, and participating in the formation of belief systems. In this context, communities are built around different topics of interaction and polarized sub-groups often emerge by clustering different opinions and points of view. Such polarized sub-groups can be tracked and monitored over time in an automatic way and the analysis of their interactions is interesting to shed light on the human social behavior. Even though many studies have been devoted to understand different aspects of the social network structure and its function, such as, community structure (For10), information spreading (BRMA12), information seeking (KLPM10), link prediction (LNK07), etc., much less work is available on analyzing online discussions, user opinion and public debates. In this doctoral dissertation we analyze the concept of polarization by looking at interactions among users in different Online Social Networks. Polarization is a social process whereby a social group is divided into sub-communities discussing different topics and having different opinions, goals and viewpoints, often conflicting and contrasting (Sun02; Ise86). We are interested in studying how and to what extend it is possible to extract information about polarized communities by automatically processing the data about interactions created in Online Social Networks. We present the state of the art and we propose a novel detecting method which allows to identify polarized groups, track them and monitor the topic evolution in the discussion among users of an OSN over time by classifing the keywords used in the messages exchanged. We show that it improves the state of the art and we describe case studies conducted particularly on Twitter (CLOP16; CGGL17). The benefits in understanding user opinions are detailed in the first chapters. Moreover, we use the proposed methodology and alternatives in different application contexts: misinformation (BCD+14a; BCD+14b; BCD+15), politics (CLOP16; CLOP15; CLO+15), social behaviors (CALS16a; CALS16b), and migrations (CLM+16). A further application of opinion mining is the task of predicting user behavior. We discuss the limitations and the challenges related to this research area by looking at the context of political elections and by digging into a case study of electoral prediction. We believe that the analysis of polarized communities is OSNs can be used to predict collective social behavior, but major improvements in the field can be achieved by integrating several sources of information, such as traditional surveys, multiple Online Social Networks, demographic data, historical information, events, cyber-physical data. Therefore, polarization is integrated in a framework of analysis with other dimensions (time, location) to explore social phenomena from a social media perspective. In particular, we look at the possibility to understand European perception of the political refugees’ crises by mining OSN data. The concept of polarization is related to that of controversy. Controversy describes the interaction among two or more opponent polarized communities that discuss together, often with heated tones. For some highly controversial topics (e.g., politics, religion, ethics) even though users prefer to get informed though polarized content originated in the communities they belong to, they like to share their affiliations, believes, ideals, convictions with external users in order to persuade them in joining their belief system or supporting, criticizing an event, a group, a party or a specific person. Highly polarization does not always imply controversy and vice versa. We describe the recent literature about controversy detection and we propose a machine learning approach which takes into account features related to the social network and to conversational interaction patterns. The model is able to identify controversy in a conversation without any feature related to the content of the interaction. The features are deeply analyzed and the accuracy of the model is discussed. We finally explore two opposite situations. The first is the formation of echo chambers, where a user gets informed and gives opinions in a self-contained group, whose members share a similar point of view. By analyzing communities in Facebook which consume news from scientific pages and from pages focused on conspiracy theories we confirm the hypothesis of cognitive closure of the users, weakening the idea of Social Media as a space for democratic collective intelligence. The second is the presence of deviant communities. Those are communities that emerge around what are usually referred to as deviant behaviors (CM15), conducts that are commonly considered inappropriate because they violate society’s norms or moral standards. An example of deviant behavior is the pornography consumption, that is the focus of our examination looking at content dissemination in Online Social Networks. Deviant communities are commonly considered segregated but we show that instead their content might spread far away in the Online Social Network. We analyze both situations with real case studies using Facebook, Flickr, and Tumblr data. Our work is an initial study of opinion polarization on Online Social Networks with some in-depth analyses of specific topical user communities. It brings novel contributions in: i) characterizing communities through the perspective of user polarization; ii) proposing a novel method to classify polarized users and topic evolution over time; iii) understanding user behavior from a social media perspective; iv) integrating polarization with other variables (time, space) with the purpose of analyzing a social phenomenon; v) defining controversy and how to detect it regardless of the content; vi) describing how people aggregate and share information in various contexts. Different topical communities and several OSNs are described in the dissertation, providing a general overview of the investigation field and proposing contributions to the discussion and solutions. Our research questions are part of a broader research area which is called Computational Social Science. This new discipline - which is the frame of our thesis - is a new approach to social studies by mean of novel large-scale computational tools, merging Social Science with Computer Science and Machine Learning.
| Item Type: | IMT PhD Thesis |
|---|---|
| Subjects: | Q Science > QA Mathematics > QA75 Electronic computers. Computer science |
| PhD Course: | Computer Decision and System Science |
| Identification Number: | https://doi.org/10.6092/imtlucca/e-theses/204 |
| NBN Number: | urn:nbn:it:imtlucca-27232 |
| Date Deposited: | 22 Mar 2017 11:46 |
| URI: | http://e-theses.imtlucca.it/id/eprint/204 |
Actions (login required, only for staff repository)
 |
View Item |
