User Behavior Analysis And Research Based On Web Log

With the rapid development of Internet, the amount of web users increased sharply., as well as the increase of the online shopper proportion,leading to the network users' behavior becoming more and more various and complex. Nowadays, although many scholars have studied the research direction of Web mining technology and user behavior analysis, there are still many problems. Especially under the e-commerce scenarios, as the basis of user analysis, the traditional session identification algorithm is still using static method, which cause coarse granularity and low accuracy. In addition, due to the neglection of user purchase behavior, the recommender system of many e-business platform always produce a lot of duplicate,outdated items recommendation which cause poor user experience. These problems above need to be solved seriously, thus, this paper choose the e-commerce vertical field, focusing on the analysis and study of e-commerce platform user behavior, in order to construct a User Behavior Analysis System based on Web Log.The main research and contributions of this paper are as follows:Firstly, on the basis of deeply understanding of the type and data characteristics of Web log , as well as deeply research and understanding of the Network user behavior, this article concluded the characteristics of user behavior including strong concealment, strong initiative and complex features, and this article summarized the Web log mining applications in e-commerce, social media, search direction engine, game operation, 020,P2P areas. Secondly, based on the research result above and on the situation of e-commence application, a hybrid Session Identification Algorithm is proposed, which solves the problem of lack of flexibility and low recognition accuracy of traditional Session Identification Algorithm.Thirdly, this article took fully use of machine learning algorithm and combined Kmeans with GMM Clustering Algorithm to realize a Two-stage Clustering Algorithm, and experimental result shows that the accuracy of this clustering algorithm results are close to the results of GMM and this algorithm shortens the experimental time of 15%-18% than using GMM,which achieves the complementary advantages of the above two algorithms. Finally, this article made deep research of Collaborative Filtering and Content-based Recommender Algorithm, and make analysis of Taobao logs, a recommender system model based on Commodity-Purchase-Cycle has been proposed creatively, whose finally Taobao Item alternative set has less repeat recommendations, and the experimental result shows that the accuracy of this algorithm model has increased from 10% to 15% than the state of art ItemCF.Based on the research contents and contributions above, this paper constructed and realized a User Behavior Analysis System based on Web Log. This system can label the log records acording to user needs, and then realized the multi-dimensional statistical analysis functions and user data mining functions, finally exported user portraits, which can help e-commence businessman or content provider to understand their users deeply, as well as gain huge business value through precision marketing or precision recommendation etc.