The Research On Recommendation Algorithm Based On User Interest

With the constant popularization of information technology and the continuous innovation of Internet technology, the problem of information overload affects our life more and more widely. Recommendation system has become another powerful weapon to deal with information overload problem after the search engine. Nowadays, traditional recommendation has been a great success in many fields, but with the change of user’s demand, traditional recommendation system can not generate the recommended drawbacks for user groups gradually revealed, so group recommendation system has emerged. Group recommendation system can provide recommendation service for a group of multiple users, and has gradually become one of the hottest spots in the domain of recommendation system.The drifting of user interests wasn’t taken into account in some traditional recommendation algorithms and group recommendation algorithms. There are also some deficiencies when the recommendation’s timeliness is considered. In view of these problems, this paper analyzes the influence of user interest change on each algorithm based on the research of traditional recommendation algorithm and group recommendation algorithm. And in this paper, an improved collaborative filtering algorithm adapting to the change of user interest is proposed in traditional recommendation and a group recommendation algorithm based on users’characteristics and user interest is proposed in group recommendation.The improved collaborative filtering algorithm adapting to the change of user interest which the similarity metric was improved, according to the characteristic that user interests change with time passing by. And an enhanced time decay model was also proposed to measure the estimated value. The concept drifting problem of user interests is solved and the timeliness of recommendation algorithm is also taken into account in this paper. In the simulation experiment, predictive scoring accuracy and TopN recommendation accuracy are compared among the proposed algorithm, UserCF-BP, UserCF-BE,TCNCF, and PTCF algorithm in different data sets. The experimental results show that the proposed algorithm in this paper is significantly improved recommendation accuracy and has a better adaptability.The group recommendation algorithm based on users’characteristics and user interest which the clustering algorithm was used for group detection based on the users’ characteristics. And also a collaborative filtering algorithm with punishment time was proposed, which was used to estimate users’ preference in the group. And, in view of the limitations of the existing preference aggregation strategy, proportion preference aggregation strategy was proposed, which combined various existing preference aggregation strategies to make up for their deficiencies. Finally, the score aggregation method and proportion aggregation strategy were used to aggregate each user’s preference in the group and get the group’s prediction score and at last generate the recommended results for the group. In the simulation experiment, the effect of the user’s characteristic value and the number of groups on the algorithm is observed. Meanwhile after comparing with BaseGRA, ImprovedGRA and GRAU, the experimental results show that the algorithm proposed in this paper can effectively improve the accuracy of recommendation.