Discovering Reliable Communities In Uncertain Graphs

Download or read book Discovering Reliable Communities In Uncertain Graphs written by Lin Liu and published by . This book was released on 2015 with total page 108 pages. Available in PDF, EPUB and Kindle. Book excerpt: Due to its ubiquity, graph data attracts increasing attention from data mining research community and industrial enterprises, and its application includes social networks (e.g., Facebook, LinkedIn, Myspace), biological networks (e.g., Protein Interaction Network), communication networks (e.g., Peer-to-Peer Network, Wireless ad hoc Network), traffic networks, to name but a few. Not until recently do people start realizing the inherent uncertainty of these applications. For example, due to the high throughput experiment methods and imperfect computational models, the false negative and false positive for detecting protein interactions are very high. In social networks, it is error-prone to model personal relationships as binaries for the imprecise information collection instruments, such as anonymous communication, self-reporting log. Therefore, researchers start interpreting these graph data from the probabilistic perspective. The solutions to many interesting uncertain graph problems are developed upon reliability, the most fundamental concept in uncertain setting and the counterpart of connectivity concept for deterministic graphs. Our work focuses on applying reliability concept to discover reliable community structures from uncertain graphs in different scenarios. First, highly reliable subgraph discovery problem attempts to identify all induced subgraphs for which the probability of connectivity being maintained under uncertainty is at least a user-given threshold. To solve this problem we propose a novel sampling scheme, which transforms the core mining task into a frequent cohesive set mining problem for a set of deterministic graphs. Such transformation enables the development of an efficient two-stage approach which combines a novel peeling technique for maximal set discovery with depth-first search for further enumeration. Second, relevant reliable subgraph mining problem aims to find a subgraph that contains a user-given set of vertices (seed set) and maximizes the lowest pairwise connectivity within this subgraph. To solve this problem, we propose both "adaptive'' deterministic algorithm and "adaptive'' randomized methods: a branch-and-bound method that iteratively enumerates possible candidates while using obtained reliability threshold to prune unpromising candidates on the fly, and an adaptive Metropolis-Hastings sampling method that avoids jumping into the unpromising states based on its search history. Third, uncertain graph clustering problem based on a generalized reliability measurement formulated from information theoretical perspective, aims to partition an uncertain graph into a set of subgraphs, each of which is not likely to be disconnected in the context of different instantiations of uncertain graphs. After successfully unifying uncertain graph clustering problem and graph (possible world) coding problem, a novel coded K-means algorithm is proposed.