Back to search

Democritus University of Thrace (DUTH)

Algorithmic analysis of user behavior in social media

Abstract

dc:description

Social networks are ubiquitous in our lives, the communities and the societies we participate in. Recently, the emergence of online social networks has triggered a major milestone in our interaction with our social circles. Online social networks have dramatically changed our social interactions and the way we seek information in the digital, interconnected world. Online users can interact with their family, their friends, or other users almost instantaneously, share their opinions and beliefs, and comment or react to other users' content. For this reason, online social networks play an increasingly significant role noawdays in both research and commercial applications. Naturally, most leading commercial entities are providing social connectivity services and compete for the users' online time, while researchers study a wide variety of components existing in social networks and online social networks, such as the behavior of users or utilizing the knowledge present in these massive databases to solve real problems.In the context of structural analysis, social networks can be naturally thought as a collection of nodes that represent the entities that participate in them and a collection of links among them that represent their relationships. As a result, the activities performed by online users can also be thought as structural in more than one ways. One natural example would be the decisions of users to connect to others users based on real life social relationships or based on the content they choose to expose themselves to. A significant amount of research focuses on this structural data exclusively, with the most common being the direct friendship network among individuals, as a means of understanding how opinions, beliefs or other content propagates through the network. In addition, this knowledge and understanding can then be utilized in an attempt to solve real world problems that often occur within the context of online social networks. These concepts refer to social network analysis, which is the study of social networks by using graph theory as the underlying mechanism. The aforementioned tasks and problems, however, are sometimes challenging to address as they are based on studying the complex nature of human behavior in the online world.In the work presented in this thesis we use both established and novel structural methodology social network analysis in order to recognize the online behavior of users and utilize the observations in order to attempt to solve problems that commonly arise in online social networks. Our assumption, that is confirmed -to an extent- through our analysis, is that users will behave consistently with their interests, beliefs or tastes. This phenomenon gives rise to patterns and structure to real social networks as users have the tendency to create connections with other users or other entities based on criteria of interest, taste or exposure. Our methods are exclusively structural, i.e. they rely on the links formed in an explicit or implicit network between the relevant entities, which may not be necessarily formed among real users. In summary, our analysis is twofold: (1) structural analysis and (2) based on the expected behavior of users.We focus our applications in problems that commonly arise in online social networks and are of particular interest to the context of study in this thesis: Greece. The first application being discussed is political affinity and, more specifically, applied to the multiparty, complex, political scene of Greece, as imprinted in social media. In this scenario, we apply methods in order to uncover the political affinity of important nodes of interest: the members of the Greek parliament (MPs) and the most popular news media active on social media. This topic is relevant as it is related to fundamental issues in online social networks, such as the reliability or bias in news stories. Another application discussed in this thesis is recommender systems in another area of important in Greece, which is tourism. We attempt to capitalize on the assumption that users will group related points of interest (POIs) together in order to infer the similarities among the POIs and power a recommendation system based on these similarities. Finally, we also study the preferential attachment mechanism, a fundamental mechanism that can explain certain properties of real social networks with respect to the decision of nodes to connect to other nodes. Here, the preferential attachment mechanism is studied in relation to the random sampling problem and a strictly correct and efficient implementation of the mechanism is developed as a growing random graph generator.We have managed to utilize structural data from online social networks in order to extract useful knowledge using social network analysis methods. The knowledge is originating from the simple observation that the behavior of online users is consistent with their interests and beliefs, a phenomenon that creates structural patterns in the network. Our multi-perspective evaluations throughout this thesis supports this assumption. In particular, for the application of political affinity extraction, we managed to classify the vertices to their known political parties with surprising accuracy and arranged them into the left to right political axis using the minimum linear arrangement problem, whose application is novel in the context of social network analysis. For the case of tourism and the recommender system, we showed that there exists significant useful information in user defined point of interest lists. Our recommendation system achieved high effectiveness with respect to multiple evaluation scenarios and firmly surpassed the popularity baselines that are often considered difficult to exceed. The datasets and supplementary material in these works, such as the surveys, are new. We finally developed an algorithm that imprints the preferential attachment mechanism accurately and efficiently into the Barabási-Albert model and proved both its correctness and asymptotic performance.

Degree

thesis:*
Grantor dc:publisher
Democritus University of Thrace (DUTH)
Year dc:date
2022

Author and committee

dc:creator, dc:contributor.*
Authors dc:creator
  • Stamatelatos, Georgios
  • Σταμετελάτος, Γεώργιος

Subjects

dc:subject × 16

Rights

Language dc:language
eng

Identifiers

dc:identifier.*
Identifier
10.12681/eadd/51007

Chain of custody

source
Harvested from
Greek National Archive of PhD Theses
Base URL
phdtheses.ekt.gr/eadd_oai/request
Last updated
2026-11-18
Source record
OAI-PMH GetRecord
citation

Stamatelatos, Georgios; Σταμετελάτος, Γεώργιος. Algorithmic analysis of user behavior in social media. Democritus University of Thrace (DUTH), 2022. http://hdl.handle.net/10442/hedi/51007