Research On Personalized Recommendation Based On Web Log Mining

With the rapid development about Internet, Web is used in our daily life、study、work and entertainment. Web can be regard as a huge information collection station. Itstores the all kinds of information that people need. Nowadays is a era which hassufficient information and poor knowledge. Due to various kinds of information, itmakes that the user lost the direction in the huge information. The personalizedrecommendation based on Web log mining can resolve this problem.The core of personalized recommendation is the recommendation method. Wefind that Markov model is convenient and easy prediction. It is an extensive statisticalmodel. In recent years, it began to be used with a web prediction and predictivemodels are more suitable as an intelligent recommendation system. Markov chain inWeb log mining’s personalized recommendation is researched in this thesis and themain work as follows:Firstly, the thesis study on the basic concepts and method about Web log miningand personalized recommendation. The main research focuses on the pretreatment ofWeb log mining technology and design algorithm. Based on Markov prediction model,the traditional Markov prediction model、mixed Markov prediction model、ThomasMarkov chain model is the build process and forecasting method is studied.Secondly，the thesis proposes a new prediction model based on Markov chain.The algorithm uses the hybrid tree structure instead of the traditional transfer matrix.It can be in progress by multiple orders forecast. Users’ frequent paths are behalf ofthe users’ characteristics. Through comparing the frequent path similarity betweenuser classifications, reduce the complexity of the user classification process.Introduction of Web page clustering thought, not only further compressed storagespace, and make the higher order sequence has high compatibility, therecommendation results more accurate.Thirdly, the improved algorithm used in designing the personalizedrecommendation of prototype system based on Web log mining. The prototype systemincludes offline and online. Subdivided into pre-processing module, pattern miningmodule, and the recommended module. And describes in detail the function of eachmodule, and the entire prototype system workflow. By the way, it describes in detailthe function of each module, and the personalized recommendation generation process.This thesis made a thorough study about two major problems--Web log miningand personalized recommendation,improved mining algorithm based on Markovprediction model,balance timeliness and accuracy of the recommended better.Andgive a prototype system for personalized recommendation in the end.I believe that thisstudy has a certain reference significance to website for the further realization ofpersonalized recommendation.