Abstract
dc:descriptionThis doctoral dissertation presents COTILES, a novel algorithm for simultaneous detection of communityand outliers in evolving social graphs that takes advantage of both structural and content-based information. Traditional community detection methods often rely solely on structural properties, potentially overlooking important aspects related to the content. COTILES addresses this limitation by integrating content constraints, ensuring the formation of communities that are structurally cohesive and thematically consistent. Key contributions of this work include the development of the COTILES algorithm, which extends the TILES algorithm to incorporate content-based constraints, and the introduction of label management mechanisms to maintain the relevance of community content over time. We also propose severalvariations of COTILES, including Strict COTILES, COTILES with Enhanced Content (EC), and a hybrid approach, each tailored to different application requirements. These variations allow flexibility in balancing topological and semantic features, making the algorithm adaptable to various contexts andneeds. Additionally, a significant part of the Thesis consists of extending COTILES to detect outliers by assigning an outlier score to each node, based on both structural and content criteria. This scorehelps identify nodes with deviant behavior and can be used for future behavior predictions. The outlierdetection mechanism is crucial to identify anomalies in social networks, which can have significant implications for security, fraud detection, and understanding user behavior.Extensive experiments using real-world datasets, such as StackExchange and MovieLens, demonstrate that COTILES outperforms traditional methods in forming more cohesive and content-focused communities. The results show that COTILES achieves better content stability within communitiesover time, reduces the percentage of highly overlapping communities, and provides a comprehensive measure of outlierness. Furthermore, experimental results validate that enforcing content constraints incurs minimal computational overhead; by maintaining a small, dynamic semantic footprint, COTILESpreserves the near-linear time complexity required for streaming environments. The empirical evaluation highlights the algorithm’s ability to adapt to the dynamic nature of social networks, ensuring that communities remain relevant and cohesive as the network evolves. The contributions of this dissertation have significant implications for various domains, including marketing, social network analysis, and security. In marketing, understanding community structures can lead to more effective targeting and engagement strategies. In social network analysis, the ability todetect and analyze evolving communities provides deeper insight into user interactions and influence patterns. In security, identifying outliers can help in early detection of fraudulent activities or security breaches. Future research directions include incorporating additional content types, such as multimedia data, to enhance thematic analysis of communities. Improving the scalability and efficiency of the algorithm is also a priority given the increasing size and complexity of social networks. Extending the methodologyto other types of networks, such as biological or transportation networks, could provide valuable insights and applications in those domains. Additionally, enhancing the algorithm to support real-time analysis and detection of communities and outliers would be particularly useful for applications requiringimmediate insights and actions. This work provides a robust framework for analyzing dynamic social networks, providing valuable insights into community evolution and outlier behavior. The COTILES algorithm and its variations represent a significant advancement in the field of social network analysis, addressing the limitationsof traditional methods by incorporating structural and content-based information. The findings of this dissertation pave the way for future research and practical applications, contributing to a deeper understanding of the complex dynamics within social networks.
Degree
thesis:*- Grantor dc:publisher
- Πανεπιστήμιο Μακεδονίας
- Year dc:date
- 2026
Author and committee
dc:creator, dc:contributor.*- Authors dc:creator
-
- Sachpenderis, Nikolaos
- Σαχπενδέρης, Νικόλαος
Subjects
dc:subject × 14- Εντοπισμός κοινοτήτων
- Εντοπισμός ακραίων κόμβων
- Δυναμικά κοινωνικά δίκτυα
- Community detection
- Outlier detection
- Dynamic social networks
- Επιστήμη Ηλεκτρονικών Υπολογιστών και Πληροφορική
- Φυσικές Επιστήμες
- Πληροφοριακά συστήματα
- Επιστήμη ηλεκτρονικών υπολογιστών, θεωρία και μέθοδοι
- Computer and Information Sciences
- Natural Sciences
- Information Systems
- Computer science, theory and methods
Rights
- Language dc:language
- eng
Identifiers
dc:identifier.*- Identifier
- 10.12681/eadd/62745