Mining interesting sub graphs using big data platforms

Abstract

The goal of graph mining is to extract interesting sub graphs from a single large graph (e.g., a social network), or from a database of many graphs.This thesis has modified the FSM-H algorithm (algorithm for frequent sub graph mining) and devised a new algorithm for weighted graph mining called WSM-H (W stands for weight).The proposed algorithm for weighted graph mining is applied to the graphs which have a weight attached to their edges.By defining two weight notions real weight and upper bound weight for pattern (sub graph), this algorithm mines sub graphs on the basis of weight notion and hence take out relevant patterns out of the large sets of graphs.Also in weighted graph mining, Anti-Monotonic property is not followed (unlike frequent sub graph mining).The algorithm comes up with weight notions that follow anti-monotonic property and hence tries to reduce the search space for mining relevant sub-graphs. Further we have optimized our algorithm by using Bloom Filters.Using Bloom Filter has reduced the key value pairs that are put into the Reduce Function of MapReduce Framework.Bloom Filter has been widely used in item set Mining for optimizing various algorithms. Bloom filter essentially tells whether a pattern is relevant or irrelevant by using set belonging property. This property of the bloom filter is utilized in our work to reduce the time taken by the algorithm.Since MapReduce is becoming the de-facto paradigm for computation on massive data, an efficient weighted graph mining algorithm on this paradigm is of huge demand.WSM-H is one such Algorithm in this domain. It uses iterative MapReduce based framework.Since the algorithm uses MapReduce it is capable of handling large graphs and set containing huge number of graphs.This also makes the algorithm robust