Research And Application Of Rating Prediction In Recommendation System Based Cloud Model

With the rapid development of the information age,massive amounts of user behavior data have created tremendous wealth in recent years.Analyzing user behavior has also become one of the most valuable research.In e-commerce and the Internet,the recommendation system aims to help users quickly and accurately find the part that is useful to themselves from massive information.It benefits from the continuous research on recommended technologies in the academic community,the recommendation system also evolved from the original collaborative filtering algorithm to an independent discipline that integrates technologies from various industries.Under the current recommended technology background,researchers have put more eyes on improving the performance of the recommendation system.Researching user behavior data to improve the accuracy of prediction is a mainstream research direction,and at the same time,it is also meaningful to conduct in-depth research on the problems faced by the recommendation system.Such as the user rating subjective problem,the existence of subjectivity leads to the disunity of the rating standards,which can not fully represent the user's interest in the object,and the existence of data sparsity also causes some difficulties for the mining of the user's interest.Aiming at such problems,the main research contents and innovations of the paper are as follows:1.User rating behavior data is studied in this paper,aims at the user rating subjective problem in the recommendation system,a recommendation method based on user behavior is designed by combining traditional collaborative filtering algorithms.Firstly,aiming at the subjective of user rating,we introduce and optimize the cloud model theory.Then,a method to generate rating standard by synthetical cloud model and transform user rating under the standard is proposed.Then,user clustering is further implemented to find similar groups of target users on this basis.2.To deal the problem of inaccurate score prediction caused by data sparseness,data dimension reduction and target user location are achieved by introducing cloud membership degree.And taking into account that user rating can be affected by their social relationship,we try to learn two rating prediction models by respectively using social relationships and similar groups.Finally,the rating value is obtained by using gauss transform to combine the two prediction models.Finally,this paper uses the real data set to experiment with the proposed method.Experimental results show that our method not only overcomes subjectivity of user rating,but also alleviates the poor accuracy caused by rating sparsity problem in traditional rating prediction methods.