Research On Collaborative Filtering Recommendation Algorithm Based On CLSVSM

Nowadays,the situation of Internet "information overload" and Internet users "information trek" is becoming more and more serious.In order to deal with these problems,the recommendation system has been widely applied in many fields such as social network,e-commerce and personalized entertainment due to its function of filtering information.Collaborative filtering recommendation is the most widely studied and applied algorithm in the recommendation systems.Its core is the calculation of similarity between users or items.However,the calculation accuracy is often weakened by the sparseness of rating matrix,which then leads to the degradation of the recommending quality.To address this problem,this paper proposes a new rating prediction method based on traditional Collaborative Filtering and Co-occurrence Latent Semantic Vector Space Model(CLSVSM).By applying the Co-occurrence theory into recommender systems,the latent co-occurrence relationship is mined out of the rating matrix,the missing units are then estimated by co-occurrence intensity.Thus,the impact from sparseness of rating matrix is reduced,and quality of recommending is improved.This article first through CLSVSM project and user for binary and ternary co-occurrence analysis,get items(users)the co-occurrence matrix and co-occurrence relative intensity matrix,then using items(users),the potential of the co-occurrence relation between two methods through the weighted average and maximum rating for the absence of the original score matrix unit estimates,the final score matrix to obtained by using new based on projects and user-based collaborative filtering recommendation,on this basis to get the collaborative filtering recommendation algorithm based on project co-occurrence and the collaborative filtering recommendation algorithm based on user concurrence.In order to verify the validity of the theory,this paper uses two evaluation indexes,RMSE and MAE,to analyze and compare the improved algorithm.Based on the experimental results,the weighted co-occurrence and the maximum co-occurrence are compared.Finally,the paper presents the results visually.Experimental results show that the collaborative filtering recommendation algorithm based on items co-occurrence and the collaborative filtering recommendation algorithm based on users co-occurrence than traditional collaborative filtering recommendation algorithm has more accurate recommendations as a result,the description is based on the co-occurrence relationship to calculate the similarity between film(user)more accurately,effectively reduced the impact of recommended results data sparse,significantly improve the quality of the recommendation.In addition,we found that the weighted co-occurrence method is more suitable for the recommendation algorithm based on binary co-occurrence,and the maximum co-occurrence is more suitable for the recommendation algorithm based on ternary co-occurrence.