Research On User Cold Start Based On Collaborative Filtering Recommendation System

Collaborative filtering recommendation system has been widely applied to many fields such as e-commerce websites,which can effectively solve the problem of "information overload".However,with the increasing number of new users and projects,the recommender system is faced with the challenge of cold start,which seriously affects the recommender system's recommendation quality,thus reducing the users' trust in the system and affecting the economic interests of merchants.Therefore,how to solve the cold start problem of recommender system and improve the recommender quality has become a hot issue worth studying.Based on the user cold start problem,this paper conducts research from two aspects of non-pure cold start and pure cold start,and is committed to designing collaborative filtering recommendation algorithm that is not affected by cold start problem.To solve the impure cold start problem,a cold start recommendation algorithm CSRA-FH based on fusion similarity and hierarchical clustering was proposed.First of all,in view of the shortcomings of traditional similarity searching for near neighbors,a fusion similarity calculation method is constructed through in-depth mining of demographic information,user rating information and item type information,which overcomes the limitation of searching for near neighbors from the perspective of user relationship.Secondly,as the number of users increases,the computation amount of the algorithm increases,the hierarchical clustering algorithm is used to quickly obtain the initial neighbor user set of the cold-start user,so as to reduce the time complexity of the algorithm.Finally,a comparative experiment is conducted on the Movielens dataset with the existing recommendation algorithms.The experimental results show that the CSRA-FH algorithm has improved the recommendation accuracy compared with other algorithms when facing the impure cold start problem,and can reduce the impact of impure cold start problem to a certain extent.A collaborative recommendation algorithm CRA?WSODT based on weighted Slope One and double trust is proposed to solve the cold-start problem.Firstly,the weighted Slope One algorithm is used to predict and fill the scores of normal users,and then the average value is used to fill the scores of cold users to a certain extent,so as to reduce the data sparsity globally.Secondly,based on the double constraints of user trust relationship and user rating similarity,the original user trust relationship is deconstructed.Then,based on the user trust matrix after deconstruction and the user rating matrix after padding,and considering the explicit trust relationship and implicit trust relationship,a user dual trust computing model is established.Finally,the trust model and the scoring matrix are integrated into the matrix decomposition model,and the cold users' prediction score on the target project is obtained through continuous iterative optimization.Finally,a comparative experiment is conducted on Film Trust dataset and Ciao dataset with the existing recommendation algorithms.Experimental results show that compared with other algorithms,CRA?WSODT has a higher recommendation accuracy when facing the cold-start problem,and can reduce the impact of cold-start problem to a certain extent.