Research On Correlation Rules Mining Algorithm Based On Matrix

The use of only support-confidence framework to mine association results in the generation of a large number of rules, most of which are uninteresting. Unfortunately, this is especially true when mining at low support thresholds or mining for long patterns. This has been one of the major bottlenecks of association rule mining. To tackle this weakness, a correlation measure can be used to augment the support-confidence framework for association rules. This leads to correlation rules. The added measure substantially reduces the number of rules generated, leads to the discovery of more meaningful rules, and at the same time detects real correlation between the rules. Therefore, this thesis focuses on correlation rules mining problem, and the main research contents include:This thesis analysis the advantages and disadvantages of5correlation measures, and uses Pearson’s correlation coefficient for mining correlation rules, and also discusses the typical correlation rules mining algorithm Taper, and proposes an improved algorithm TaperC for mining frequent correlated item pairs. The algorithm is composed of4steps:Boolean matrix construction, matrix compression, frequent item pairs extraction and correlated items pairs extraction. The advantage of algorithm lies in that, employing matrix method, only scanning the database once and reducing candidate test time by using the Boolean AND operator. Experimental results show that TaperC perform better than Taper.In addition, this thesis studies in depth on typical graph correlation rules mining algorithm FCP-Miner and proposes an improved algorithm FCSP. The algorithm inherits TaperC’s matrix method to discover all frequent subgraph pairs, and then calculates Pearson’s correlation coefficients to discover all correlated subgraph pairs. Experiment results indicate that FCSP algorithm is more effective and efficient than FCP-Miner.