Association Rule Based Personalized Recommendation Research

Personalized recommendation system was one of effective tools which could solve the problem of information overload and amount of research about it increased year by year. Many recommendation algorithms which needed lots of customer information conflicted with the fact that customer privacy protection was paid more and more attention. Thanks to advantage of easy, audio-visual and off-time computing, association analysis method was always applied to commercial recommendation system especially in the area of non-structural environment.However, association rule tools always generated lots of rules. How to get the final recommendation result was important research task. The method to directly select best rule with ignorance of background would maybe make the result that the real satisfying recommendation was eliminated for the low support and confidence. Customer feedback information was easy-access and valuable domain knowledge.Two research directions for how to get the final recommendation result:the first one is still to select best rule, the second one is aggregating rule information as much as possible. This paper gives two algorithms to solve two problems.(1)Aiming at the problem that association rule recommendation with customer score information came across the problem that there were multiple evaluation criteria for rules and dimension of criteria were different, this paper proposed the analytic hierarchy process based association rule recommendation algorithm, which combined qualitative and quantitative method and gave an effective solution for best rule selection.(2)Aiming at the situation that only contains commodity’s name and score information, this paper proposed a behavior and score similarity based association rule group recommendation algorithm, in which the rule with its scores was regarded as an expert, the experts with same conclusion were grouped together and the expert weights were calculated based on both behavior similarity and score similarity. A better recommendation suggestion was reached by aggregating the recommendation opinions of the experts.(3)After two algorithms, this paper give the experimental annlysis with data set of Grouplens Research to analysis the rate of precision and satisfaction and give experimental analysis chart.With the usage of AHP, behavior similarity, score similarity and group decision-making, this paper solve two problems in the association rule recommendation algorithm. As a result, the rate of precision and satisfaction are improved, and a new way is found for recommendation system future research.