A Collaborative Filtering Recommendation Algorithm Based On Probabilistic Model

With the arrival of the big data era,the problem of information overload is becoming more and more prominent.The role of recommender systems is becoming more and more important,and its application is more and more extensive,which further promotes the research and development of recommendation technology.Recommender system is able to help users to find interesting information from a large amount of resources,and then provide personalized services,so as to effectively solve the problem of information overload.At present,recommender systems are widely applied in the field of e-commerce,and users improve the requirements of projects and services.The recommendation system is facing the problem of accuracy of recommendation results.This thesis,the study of ranking-oriented collaborative filtering recommendation algorithm based on probabilistic model,is aimed at solving the serious problem existed in the current situation that the recommendation system's accuracy of the results,the improved listwise collaborative filtering recommendation algorithm which considers user-item factor,time factor and user factor is aim at improving the accuracy of recommendations.As for the user-item type characteristic,this thesis proposed the project to construct characteristic information of the users,calculating the similarity of the user's item type rating which is based on Plackett-luce model,the user's similarity is the combination of the similarity of the user's item rating and the similarity of the user's item type rating.As for the time characteristic,this thesis proposed the use of the attenuation characteristics of the Logistic equation,adding user's time of interest to item in the calculation of the probability of top-k permutation.As for the user attributes,this thesis proposed the weighted of user attributes similarity,which is adding user attribute factors in the user similarity calculation.And using recommendation of features,in order to improve the accuracy of recommendation,as far as possible to reduce the impact caused by the sparsity of recommendation system.Finally,we conducted a extensive experiments on the MovieLens data set in comparison with the state-of-the-art approaches to demonstrate the promise of our approaches.The experimental results show that the improved listwise collaborative filtering algorithm can improve the accuracy of the recommendation results.