Mining cohesive subgraphs is a fundamental problem in the analysis of real-world networks, with applications ranging from identifying protein complexes in biological networks to detecting communities in social networks. The k-clique, defined as a set ...
Mining cohesive subgraphs is a fundamental problem in the analysis of real-world networks, with applications ranging from identifying protein complexes in biological networks to detecting communities in social networks. The k-clique, defined as a set of k pairwise adjacent vertices, serves as the most basic building block of cohesive structures. However, listing and counting k-cliques in large graphs remain computationally challenging tasks due to the combinatorial nature of the problems.
In this thesis, we present efficient algorithms for both the exact listing and approximate counting of k-cliques. First, we address the k-clique listing problem. Existing approaches for the k-clique listing problem are built upon a recursive framework, which often suffers from repeated computations of smaller cliques. In contrast, our algorithm computes each clique only once and stores it into a novel data structure called the Induced Subgraph Trie, which allows us to retrieve the cliques efficiently. Furthermore, we propose a simple yet effective pruning technique, which further improves the running time. Extensive experiments on real-world networks demonstrate that our algorithm achieves up to two orders of magnitude speedup over the state-of-the-art algorithm while using less memory.
Second, we present an efficient Monte Carlo algorithm for estimating the number of k-cliques with theoretical accuracy guarantees. We introduce a suite of novel techniques designed to obtain a small sample space and reduce the computational cost of sample trials. Additionally, we derive a new criterion for stopping sample trials while achieving the desired accuracy guarantee. Experimental results verify that our algorithm achieves up to two orders of magnitude speedup over the state-of-the-art algorithm while maintaining accuracy guarantees.