Research On Personalized Recommendation Algorithm Based On User Behavior

With the continuous development of the Internet,the problem of data overload is becoming more and more obvious.It is difficult for users to get the part they are interested in from the massive network data.In order to solve this kind of problem,two kinds of tools,search engine and recommender system,are produced.Recommender system has more novelty and initiative,so it is more concerned by the academic and business circles.A good recommendation algorithm can accurately analyze user preferences according to the historical data generated by users,and then generate a unique recommendation list to guide users to find effective information.In order to intuitively obtain user preferences,most of the recommendation algorithms use explicit scoring to recommend,but this kind of data is generally not easy to obtain and can not guarantee the authenticity,so it often produces a sparse matrix when building the scoring matrix.In fact,when users browse the website,they will produce a lot of implicit feedback data,which are not detected by users,but can truly show the user's behavior trajectory.There are many invalid or abnormal data in the behavior data,which can not directly reflect the value in the recommendation,so it can not be well used.To solve the above problems,this paper proposes a personalized recommendation method based on user behavior,which alleviates the problem of sparse matrix in the recommendation process,and proposes a method of converting implicit feedback data into display score.The main contents of this paper are as follows(1)In order to reduce the interference caused by abnormal behavior,the rule-based method is used to filter the data and extract the normal behavior.The user's behavior is quantified,and the implicit relationship between behavior and interest is accurately mined.The general algorithm will subjectively give scores to the behavior according to the importance of the behavior,but this method is relatively rough and has poor interpretability.This paper proposes a model to measure the relationship between user interest and user behavior.It uses AHP and entropy weight method to weight user behavior,which can effectively convert implicit feedback data into display rating data and improve the accuracy of recommendation.(2)Due to the great personality differences between users,some users like to "pick and choose" and "shop around" when purchasing goods,but some users don't like to waste time when purchasing goods,and they will buy the goods they like directly.Therefore,the data reflects that there will be differences in the proportion of users' browsing and purchasing,Therefore,this paper integrates the purchase rate of different behaviors,modifies the converted score data,and uses collaborative filtering algorithm to recommend,which can increase the matching degree between similar users.This algorithm is tested on spark platform,and the experimental results show that the accuracy of the recommendation results is improved.