Research And Application On Web Data Mining Oriented Personalized Recommendation System In E-commerce

With the rapid development of information technology and the high popularity of theInternet, the Internet has played an indispensably important role in people’s lives.As a newbusiness model,E-commerce is changing people’s habits. With the continuous expansion of thescale of E-commerce sites, how to help users quickly and accurately find services and items ofinterest has become a hot research today. In this case, personalized recommendation systemscame into being.Personalized recommendation systems dig out the potential users’ interest and the currentdemand to recommend services and items to users based on the historical behavior records of theusers, using various techniques including statistical analysis and data mining techniques.Recommendation technology based on collaborative filtering is a current research hotspoton the personalized recommendation field. The technology selects the appropriate measuremethod to calculate the similarity of users or items based on the explicit or implicit ratingrecords of users, and selects users(or items) having high similarity to build user(or items) nearestneighbor set, finally, recommending items which the users among user nearest neighbor set likeand the target users haven’t bought to the target users. Therefore, building nearest neighbor setdirectly relates to the accuracy of recommended results. After in-depth research on existingproblems on the traditional collaborative filtering recommendation technology, this paperproposes two improved algorithms.The main work of this paper consists of the following aspects:(1) Deep analyze the existing shortcomings of traditional similarity measurement methods,and deep research on cloud model theory, then we introduce the idea of the cloud model and usethe similarity measurement method combined with cloud model.(2) Propose the recommendation algorithm combined the cloud model with item ratingprediction.The algorithm can effectively solve the extreme sparsity problem on the rating matrix, and overcome the shortcomings of traditional similarity measurement methods, improving therecommended accuracy. Verify the superiority of the improved algorithm via comparisonexperiments on publicly available data set MovieLens.(3) Propose the recommendation algorithm combined users neighborhood model withmatrix decomposition. The algorithm builds neighborhood model of user profiles to improve theaccuracy of recommendation for new users, and uses stochastic gradient descent on singularvalue decomposition of the rating matrix, effectively solving the problem of sparse matrices.Verify the effectiveness of the algorithm with comparison experiments on MovieLens datasets.