Research On Data Sparsity Problem In Collaborative Filtering Systems

With the rapid development of the Internet,the way to get information is gradually increasing,bring convenience to people,the mass of information also covered the people's daily life,"information overload" has become a real problem can not be ignored.How to quickly and accurately find the information needed by users in the infomation ocean has become a hot issue in the current research.In order to solve this problem,the personalized recommendation system appeared,becoming important way to get information after the search engine.The personalized recommendation system includes many types,among which the most widely used is collaborative filtering recommendation.Its advantages include simple calculation,easy implementation and so on.On the other hand,it also has some problems,such as data sparsity,cold start,and user interest fluctuation,which directly affects the accuracy of the results.Therefore,how to improve the accuracy of the recommendation has become an urgent problem to be solved.In order to solve the above problem,many representative effective recommendation algorithms have been proposed,such as Deng Ailin et al proposed a predictive score item based collaborative filtering recommendation algorithm;Sarwar et al.proposed the score matrix filling value method etc.But in the collaborative filtering algorithms do not take the user item rating matrix very sparse and traditional,between the user common rating items are scarce this situation into account,when the situation when using traditional similarity calculation method will lead to inaccurate results recommended.Aiming at the problems,this paper puts forward to combine the user's rating time to find users with similar scoring behavior,so as to improve the traditional collaborative filtering algorithm to find the nearest neighbor method,compared to the traditional user based collaborative filtering algorithm,MAE is reduced by 1.89%;An improved algorithm is proposed and the user score variance similarity calculation,thus comprehensively use user rating information to improve the similarity,the method makes mae is reduced by 1.94%;An improved algorithm is proposed based on the combination of user ratings,which makes mae is reduced by 2.19%.Through experimental results,even in sparse data,users of the common score scarce under the premise of the improved algorithm to calculate similarity between users is still relatively accurate,to improve the accuracy of recommendation.