Rule-Chain Incremental Mining Algorithm Based on Directed Graph

Abstract

Aiming to discover the rule-chains directly, a novel graph-based online incremental mining algorithm (RIOMining) is proposed. Firstly, an interrelated bitmap based on a directed graph (PAGraph) is designed to compress the rule-chain information for storage. The compressed storage of the directed-graph effectively avoids repetitious database scanning and makes meaningful rules found more conveniently. Secondly, four theorems about discovery and pruning of frequent path in the graph are proved, so as to the rule-chains could be iteratively generated by searching incremental paths, while the theorems narrow the search scope and promote the performance by properly pruning and extending paths. It effectively reduces the redundancy calculation. Comparing it with other two fast mining algorithms, DLG and FP-growth, experimental result shows that it can mine rule-chains of high quality, especially, it is of real-time accuracy, and thereby it can be widely applied to dynamic network environment.