Applied Research On Clustering And Association Algorithms In The Mobile User Behavior Analysis

Data mining technology is one of the forefront research topics in the information science field,which has widely applied in marketing, medicine, agriculture, telecom-munications and so on. With constant aggravation of the telecommunication market competition, how to implement difference marketing and service to different customers group,how to subdivide and classify the active demand that has already become the urgent needs of the telecommunications business. According to the actual situation in engineering, this paper studies the application of data mining and data warehousing technology in the mobile user behavior analysis.In this paper, it is introduced the data warehouse and data mining of the basic theory and the main data mining methods which usually are used,and analyzed the necessity and feasibility in the telecommunication enterprise.On this basis,this paper thoroughly studies the Clique algorithm in clustering algorithms and the Apriori algorithm in association algorithms, and gives the modelling process to the mobile user behavior analysis.The Clique algorithm has the grid class algorithm efficiency high merit, is extremely effective for large databases of high dimensional data clustering. But in the parallel processing, the zero data processing, reduces the denticle edge data, it have certain limitation. Therefore, for mobile user data set's characteristic,this paper discussed Clique algorithm and proposed to improve the one-dimensional Clique algorithm through the data stored in relational databases(Oracle) and sql language Experiments show that the data pre-processing results are satisfactory, run more efficient.As the association rule mining of data sets of up to GB and even TB in practical work, the traditional association mining algorithms Apriori must to reduce I/O operation and to reduce calculating the support item set. In this paper, because of the limitations of existing Apriori algorithms, the improving Apriori algorithms dynamically adjusted running in parallel is proposed. In each cycle, first, according to the rules reduce the transaction data set D, then, k-large item sets is splitted into multiple k-small item sets with Hash algorithms on k-1 ID, so it can be run on multi-machine or multi-process run in parallel. Experimental results show that this method can significantly improve the mining efficiency, making a time-consuming work on the original mass of data mining association rules can be taken in the process of multi-machine or multi-process implementation.Finally, the paper discusses how the improved algorithms are applied in the mobile phone behavior analysis system. According to the data mining design ideas, this paper gives an analysis of system design and build a new customer segmentation model of the mobile user.Take BOSS system as the actual data source, carries on the data pre-processing using the improved Clique algorithm,and using the improved Apriori algorithm, carries on the classification and the forecast to competitor's new users. The result indicated that the improvement algorithm has the good operating efficiency and the forecast effect in the practical application.