The Research And Improvement Of Weighted Association Rule Algorithm In Data Mining

Data mining is used to extract useful information from large –scale databases, even predict the tendency of the future things according to current ones. In the field of data mining, association rules is one of the main research content, which is used to indicate the rules or patterns of relationship for data item sets. The main aim of this paper is researching for weighted association rule mining algorithm based on Apriori algorithm in data mining, and applied to the electronic commerce recommendation system to verify it.At first, the basic knowledge of data mining and association rules is introduced in detail. The Apriori algorithm is analyzed, expounded and made performance analysis. For Apriori algorithm may ignore the small probability but important project, and generate too much boring association rules, Introducing weights of ideas may avoid the possibility of the important things be ignored. Weighted association rules algorithm in the general definition and model have been researched, with the k-supporting expectations as the basis of the pruning process, which can overcome the defect of weighted algorithm that can’t convergence; introducing the idea of matrix as data storage, to overcome the defects of traditional association rules algorithm, and reduce the consumption of time and space. Based on the above two points, the algorithm which is based on weighted association rule model and the algorithm of matrix, is introduced. Using matrix calculation of item sets support for peace, don’t need to scan the database, the database general scanning frequency to once; transform the weighted support calculation model, with the number of minimum weighted support instead of the process that comparing the minimum weighted support process and weighted support of each set calculated. Only using matrix operations support of new sets, and the corresponding minimum support number contrast to produce frequent k – item sets directly without generate candidate item sets, during the connecting of(k- l)- frequent item sets. The ideas of the detailed algorithm and implementation process is introduced in detail. The algorithm avoids a large number of I/O operations, significantly reduce the amount of time, had certain rules convergence than traditional weighted algorithm, through multiple sets of data instances verification. It has been proved to have stronger effectiveness.At last, the mining algorithm improved is applied to the recommendation of the commercial system, and be analyzed with the difference of the traditional recommendation system used a weighted association rules. In offline mode, the improved algorithm greatly improve the speed of the algorithm, save the time of generating weighted association rules and strengthen the friendly human-computer interaction interface. The feasibility of the algorithm is verified by experimental data,and association rules and the results mined will play a positive role in guiding to the user.