University of Crete (UOC)
Analysis of evolution, dynamics and vulnerabilities of online social networks
Abstract
dc:descriptionOnline Social Networks (OSNs) are offering an experience that goes beyond communication, news or entertainment. With a total user base that reaches the one third of the world population and an average daily engagement of three hours, OSNs have become a major phenomenon that affects our society in a variety of ways. Also OSNs have already a history of almost 30 years of constant growth, creating a sizable market that attracts considerable funding and innovation. Inline with this growth, there is a parallel increase of interest from the scientific community that attempts to study OSNs from various perspectives. Without being complete, these perspectives can be delineated according to the way the community treats an OSN as a research object. First of all, an OSN can be perceived as a complex system represented by a social graph that is continuously changing. A second perspective is as a social phenomenon that hides many dangers from which the public should be informed and protected. A final view of OSNs is as a tool, through which we can focus on some interesting trends and tendencies inherent in the public sphere. This dissertation presents some fundamental contributions in these areas and uses Twitter as a testbed for experimentation and validation. Initially, we present an effort to model the temporal evolution of the growth of the social graph. Towards this goal, we collect two datasets containing daily snapshots of the social graph, one for the early and another for the later period of Twitter. By fitting this dataset to a well-known but previously untested model, we are able to graph the evolution of Twitter for a period of 8 years. Additionally, we annotate the observed fluctuations of this growth with real events and demonstrate how efficient spam control and service robustness can affect the growth of an OSN. We proceed to study one of the most common strategies for spam propagation in OSNs. This is the deliberate mix of popular topics with spam content. By using Machine Learning methods, we show that the use of trending topics has the maximum discriminatory efficiency between spam and legit content. Also, we uncover a spam masquerading technique and we show how we can mitigate spam with simple graph analysis and computationally modest machine learning models. Finally, we delve into content analysis.Specifically, we apply a combination of Natural Language Processing techniques to infer how users express themselves during a real and turbulent electoral event. Towards this, we apply Named Entity Recognition, Volume analysis, Sarcasm detection, Sentiment analysis and Topic analysis in order to extract among other, the semantic proximities of different political parties and the temporal sentiment variation of different groups of voters
Degree
thesis:*- Grantor dc:publisher
- University of Crete (UOC)
- Year dc:date
- 2020
Author and committee
dc:creator, dc:contributor.*- Authors dc:creator
-
- Antonakaki, Despoina
- Αντωνακάκη, Δέσποινα
Subjects
dc:subject × 25- Μέσα κοινωνικής δικτύωσης
- Μηχανική μάθηση
- Αναλυση γράφων
- Επεξεργασία φυσικής γλώσσας
- Εξαγωγή σημασιολογικών σχέσεων
- Ανεπιθύμητα μηνύματα
- Πρόβλεψη εκλογικού αποτελέσματος
- Ανάλυση συναισθήματος
- Ανάλυση σαρκασμού
- OSN
- Average degree
- Sentiment analysis
- Spam analysis
- Sarcasm detection
- Graph analysis
- Evolution analysis
- NLP
- Machine learning
- Topic analysis
- Classification
- Φυσικές Επιστήμες
- Επιστήμη Ηλεκτρονικών Υπολογιστών και Πληροφορική
- Natural Sciences
- Computer and Information Sciences
Rights
dc:rights- Statement dc:rights
-
- BY_NC
- Language dc:language
- eng
Identifiers
dc:identifier.*- Identifier
- 10.12681/eadd/47236