Graphlab Geometric Multiway Recurrent Pattern Extraction Algorithm For Recurrent Pattern Generation In Distributed Graph System
A recurrent pattern is required for fact-finding analysis on graph databases and it is a subgraph that presents frequently in the graph database. Recurrent pattern extraction is a part of Data Mining where more research is going on to fulfill the current era requirements as the data size grows at a faster rate. The researcher has to focus on developing a methodology for candidate generation (with duplicate elimination) and frequency count that is mathematically efficient and policy-wise effective. In a distributed system, it is very difficult to calculate the global frequency of any pattern during the calculation phase. In a top-level parallel framework, like Pregel, MapReduce doesn’t support it. In this paper, we propose a GraphLab Geometric Multiway Recurrent Pattern Extraction (GGMRPE) algorithm that solves these issues present in existing methods. A cache consistency process arises in GraphLab, which is to be solved by the versioning method. Load balancing is done by a homogeneous partition and an equal number of ghosts. We used canonical ordering of locks to avoid the deadlock and implement the Lamport-Chandy algorithm to overcome the fault tolerance. We performed an experimental investigation with four real graph datasets with different criteria and found drastic improvement (i.e. 75% improved) with various algorithms such as PageRank, Connected Component, Triangle Counting, Collaborative Filtering, and Belief Propagation.