Research On Collaborative Filtering Recommendation Algorithm For Data Sparse Optimization

In the data age of information overload,personalized recommendation system can recommend information of interest to users through information filtering technology,so it is widely used in various fields.Among them,collaborative filtering algorithm is one of the most widely used and popular recommendation algorithms.However,collaborative filtering still has the problem of data sparsity,which seriously affects the quality of recommendation.Aiming at the data sparsity problem of collaborative filtering algorithm,this paper studies it from filling sparse scoring matrix and defining a new user similarity model.The main work accomplished in this paper is as follows:First,in order to improve the accuracy of collaborative filtering algorithm when data is sparse,this paper proposes a data-filled collaborative filtering recommendation algorithm based on auxiliary information.The main features of the algorithm are as follows:(1)When generating filling data,the user/project auxiliary information is integrated to represent the user/project characteristics,which can generate filling data for new users and new projects,and can accurately measure user/project similarity.In the fusion of user auxiliary information,this paper fuses user's basic attribute information,and predicts user's preference for project attribute based on user rating and item attribute information.In the fusion of project auxiliary information,the project attribute information,title information and content information are effectively fused.At the same time,noise reduction coder is introduced to mine the low-order dense implicit features of user/project.(2)When filling the matrix,the confidence of filling data is considered,and the influence of noise data on recommendation quality is reduced by filtering the filling data with low confidence.Tests on extremely sparse data sets show that the proposed algorithm can alleviate the problem of data sparsity and has high recommendation accuracy.Second,aiming at the problem that the traditional user similarity model can't be applied to sparse data because it relies entirely on users' common scoring items,this paper proposes a collaborative filtering algorithm based on hybrid similarity.Firstly,the algorithm introduces project attributes to calculate project similarity,and measures the impact of non-common scoring items on user similarity by combining project similarity and scoring similarity.It no longer relies on users' common scoring items,so as to alleviate the problem of data sparsity.Secondly,in order to measure user similarity comprehensively and objectively,this paper considers a common score reward factor and user credibility factor.The common score reward factor is used to measure the impact of the proportion of common items and attributes on user similarity,and the user reliability factor is used to measure whether users have unreliable scores,so as to reduce the impact of unreliable users.Finally,the proposed algorithm is compared with the algorithms based on other similarity model.The experimental results show that the proposed algorithm can be better applied to sparse data,and the recommendation accuracy is significantly improved.