Research On Methods Of User Behavior Based Personalized Recommendation Optimization

With the development of electronic commerce and the popularity of the Internet,the user oriented personalized recommendation is paid more and more attention,accurate recommendations could not only help users save a lot of time but also could help the electronic commerce websites win the users' attention,even increase sales.For different users have different backgrounds,and users want different recommendations,therefore,it is necessary to mine the hidden information from multiple perspectives,and provide more accurate personalized recommendation results.This paper explores user preferences based on user's potential behaviors and social relationships,optimize user personalized recommendations from the following aspects:Research on user behavior and relationship according to time characteristics of user behaviors and preferences,and improve the efficiency of user preference prediction;Study relationship between web topology and weight of web pages,and then optimize the page ranking for users to provide accurate recommendation;Study the negative social relations and the probabilistic characteristics of topic being recommended to users to give recommendations for inactive users who have scarce historical information,optimize fusion of negative social relations and the recommended topic probability characteristics to give accurate recommendations for inactive users.We optimize personalized recommendation by following methods:(1)proposed personalized recommendation optimization method based on behavior perception,which analyses the history of user access behaviors,establishes hidden Markov model of user behaviors and preferences,reduces the time of optimizing user parameters by clustering,and gives personalized recommendations with balancing the accuracy and time complexity.(2)proposed user recommendation optimization method based on Web topology identification of web page anomaly ranking.Based on topology of webpage we identify abnormal web page ranking,research the effect of web page topology on web page weight,and compare the weight and number of the linked into web pages to identify the phenomenon of abnormal page ranking.This work creates a fair web page ranking environment which could improve the quality of personalized recommendation.(3)proposed fusion of negative social relationships and Poisson process based recommendation optimization method for non-active users,expanded negative social relationships based on initial negative social relationships and the attenuation transfer coefficient matrix.The expectation of a topic for a non-active users is based on the constraints of negative social relations,then the probability of user satisfaction to the topic is predicted based on Poisson process.The topics with the highest probability are regarded as the final recommendation for non-active users,which could give accurate recommendations for non-active users.