Advisory Committee Chair
Da Yan
Advisory Committee Members
Zhe Jiang
Sidharth Kumar
Virginia Sisiopiku
Kai Zhao
Yang Zhou
Document Type
Dissertation
Date of Award
2023
Degree Name by School
Doctor of Philosophy (PhD) College of Arts and Sciences
Abstract
Quasi-cliques and k-plexes are dense structures with established significance in graph mining, offering flexibility and resilience to data anomalies. However, mining these structures poses computational challenges due to their NP-hardness. In this study, we address these challenges and propose solutions for efficient mining, in addition to visualization tools for tuning and analysis. For quasi-clique mining, we adapted the distributed solution (G-thinker) [1] to a sharedmemory multi-core environment, making it accessible to average users. Our solution is two orders of magnitude faster than the recent solution by [2], and it scales almost ideally with the number of CPU cores. To facilitate user interaction and parameter tuning, we created QCQ-Viewer, enabling fine-tuning of mining parameters and intuitive examination of resulting quasi-cliques. In the context of frequent subgraph pattern mining, we introduced T-FSM [3], a system for efficiently mining frequent patterns in big graphs. T-FSM ensures high concurrency, limited memory usage, and effective load balancing, incorporating the more accurate Fraction-Score frequentness measure. To enhance usability, we developed FSM-Viewer, a graphical user interface enabling users to mine frequent subgraph patterns and explore matched instances batch by batch. This thesis also addresses the k-plex mining problem from two angles: the maximum k-plex problem and the maximal k-plex problem. For the maximum k-plex problem, we summarize recent works and introduce a visualization tool for network analysis. Regarding the maximal k-plex problem, we design new algorithms and new pruning and branching rules to efficiently find maximal k-plexes. This study makes a valuable contribution by introducing efficient mining techniques for quasi-cliques and k-plexes in large graphs. Moreover, it also provides visualization tools that facilitate the exploration of significant structures and insights in various applications. As a result, this research enables the discovery of meaningful patterns and enhances our understanding of complex networks.
Recommended Citation
Khalil, Jalal, "Large-Scale Mining of Dense Subgraphs: Algorithms and Systems" (2023). All ETDs from UAB. 413.
https://digitalcommons.library.uab.edu/etd-collection/413
