CS Other Presentations

Department of Computer Science - University of Cyprus

Besides Colloquiums, the Department of Computer Science at the University of Cyprus also holds Other Presentations (Research Seminars, PhD Defenses, Short Term Courses, Demonstrations, etc.). These presentations are given by scientists who aim to present preliminary results of their research work and/or other technical material. Other Presentations serve as a forum for educating Computer Science students and related announcements are disseminated to the Department of Computer Science (i.e., the csall list):

RSS Directions

Presentations Coordinator: Demetris Zeinalipour

PhD Defense: Unsupervised Extraction and Analysis of Polarization in Large-Scale Text-based Corpora with Applications to Misinformation Detection, Mr. Demetris Paschalides (University of Cyprus, Cyprus), Tuesday, December 17, 2024, 10.00-11.00 EET.

The Department of Computer Science at the University of Cyprus cordially invites you to the PhD Defense entitled:

Unsupervised Extraction and Analysis of Polarization in Large-Scale Text-based Corpora with Applications to Misinformation Detection

Abstract:
Polarization is a growing global issue that poses serious threats to social cohesion, public trust, and the integrity of democratic institutions. Deepening divisions between individuals and groups, especially on contentious issues, have been particularly evident in recent events like elections, referendums, the COVID-19 pandemic, and geopolitical conflicts. Digital platforms have exacerbated this issue by facilitating the spread of information dysfunction—such as misinformation, disinformation, hate speech, and media manipulation. These challenges underscore the need to understand the mechanisms of polarization in order to identify effective mitigation measures. While existing polarization studies have been effective in specific contexts, such as analyzing the known segregation on the political spectrum, they often rely on controlled case studies, curated datasets, and prior knowledge. However, reliance on such methods can limit their broader applicability and may overlook the multi-level nature of polarization that spans entities, groups, and topics. This can restrict their ability to provide comprehensive insights into polarization dynamics across different domains. In this thesis, we address these challenges by proposing an unsupervised and domain-agnostic framework for the modeling, extraction, measurement, and analysis of multi-level polarization. Central to this framework is the Polarization Data Model (PDM), which conceptualizes polarization as a multi-level phenomenon occurring at the entity level (individuals and their positive or negative attitudes), group level (fellowships with shared beliefs), and topic level (conflicting attitudes towards various subjects). Represented as a knowledge graph, the PDM encapsulates entities, their interactions, fellowships, opposing groups (dipoles), topics of discussion, and the attitudes expressed. To construct the PDM, we introduce POLAR, an unsupervised pipeline designed to extract polarization knowledge from news articles. By leveraging the formal nature and rich content of news articles, POLAR processes their content to uncover how entities interact and express attitudes towards each other. It analyzes these interactions to identify fellowships of entities that share common views, and detects dipoles that represent opposing fellowships. Through this process, the framework also identifies the key topics driving these divisions and quantifies the degree of polarization present. This approach allows for the identification of polarization dynamics across multiple levels without relying on prior knowledge or predefined categories. Evaluating polarization is inherently challenging due to its subjective nature and the absence of ground truth data, which makes it difficult to validate polarization measures across different domains. To address this, we propose a Multi-level Evaluation Methodology that assesses polarization at the entity, fellowship, and topic levels. The evaluation compares the extracted polarization knowledge at each level against ground-truth data, baseline models, and state-of-the-art methods. This robust evaluation process ensures that POLAR is both reliable and effective across diverse domains. We assess the correctness of the proposed framework through case studies of Abortion, Immigration, and Gun Control. Our results show strong performance across these levels, with particularly high accuracy in fellowship alignment and effective ranking of topic polarization. In addition, we showcase the utility of our framework via a large scale study of the COVID-19 pandemic in the US. Furthermore, we explore polarization relations with other information dysfunction phenomena, particularly misinformation. Specifically, we propose PARALLAX, a framework that incorporates polarization knowledge to existing misinformation detection methods to improve their performance. PARALLAX utilizes the extracted polarization knowledge to encode content to be tested for their veracity, and combines this as a feature to existing classifiers via FlexKGNN, a novel graph neural network. We tested our methodology on three misinformation datasets, demonstrating that it achieves approximately a 15% improvement in performance over baseline classifiers.

Short Bio:
Demetris Paschalides is a Ph.D. candidate in the Department of Computer Science at the University of Cyprus and a Research Assistant at the Laboratory of Internet Computing (LInC). He earned an MSc (2018) and a BSc (2016) in Computer Science from the University of Cyprus. Demetris has worked on projects like MANDOLA and Check-It, which focus on tackling critical challenges such as hate speech, misinformation, and polarization. His research integrates advanced techniques in Data Mining, Natural Language Processing (NLP), and Machine Learning (ML). Demetris’ research has been published in IEEE/ACM venues, including WI, ASONAM, ToIT, OSNEM, and TSC.

	Other Presentations Web: https://www.cs.ucy.ac.cy/colloquium/presentations.php
	Colloquia Web: https://www.cs.ucy.ac.cy/colloquium/
	Calendar: https://www.cs.ucy.ac.cy/colloquium/schedule/cs.ucy.pres.2024.paschalides.ics