Dynamic Weighted Association Rules Mining Based On Browsing Behavior And Cluster Analysis

Association rules minning,which is used for finding the potential connection of transaction in database,is the important branch of data minning and personal recommendation.Moreover,it is the research focus that weigted association rules minning could be used for solving the problem of different importance of items in Database.This paper firstly introduces the existing algorithm of the weighted association minning.After the deep research,there are some problems of the existing algorithm as follows.First,weights of items are setted subjectively.Second,weights break the closure property of the traditional algorithm.Third,the low recall rate and precision ration of the recommendation result.For solving the above problems in the e-commercial application,the main work of this paper is as follows.1.Solve the fixed and subjective problem of weights by constructing users'interest model and bring in the attenuation mechanism of the interest.It makes weights could dynamically reflect the interest of users that users' interests of items are treated as weights of items.2.It is more effective to combine the weighted association rules with improved clique cluster algorithm which is based on data characteristics.The improved algorithm which is more effective for high demsion data makes users who have the same interest in the same cluster.This way reduces the number of mining data and improves the performance of algorithm,3.Introduce the method of transaction weight to make the algorithm have the closure property.It makes the algorithm have the closure property to regard a interest record of a user as a transaction.Finally,mine the association rules in clusters by improved FP-Growth algorithm.Compared with the existing algorithm,the improved one could have more reasonable weights and dynamically reflect users' interest change.Moreover it satisfies the closure property,namely that a frequent set's subset is frequent.After clustering users by improved clique algorithm,there is less data in cluster.By testing the improved algorithm with other similar algorithms,it is verified that it has more effective performance.