Research On Collaborative Filtering Algorithm Based On Combined Similarity And Matrix Factorization

The rapid development and popularity of the Internet has greatly enriched and facilitated people's lives,and also brought many new problems and challenges: On the one hand,with the increase in the diversity of Internet products and the complexity of system structures,and the variability of user behavior And the diversity of data types,the traditional collaborative filtering recommendation algorithm is difficult to meet the user's personalized recommendation needs in terms of recommendation speed and recommendation quality;on the other hand,the rapid development of Internet technology has seen a large number of Social platforms represented by Weibo and We Chat,and e-commerce platforms represented by Taobao and Jingdong Mall,have generated mass data centered on users.How to use this information to alleviate the problem of data sparseness and timeliness,this article focuses on the following aspects:(1)Analyze the data through the time forgetting function and user bias degree to solve the problem of user interest drift and data sparseness in collaborative filtering recommendations,and propose a collaborative filtering algorithm T-UBSM based on time function and user bias degree(Similarity calculation Method based on Time function and User Bias).Firstly,the time factor is incorporated into the user item scoring matrix through the MCM index function,and the user interest is tracked over time to solve the problem of user interest attenuation and drift.Based on this,the user bias degree is defined to improve the similarity calculation method.Finally,a new similarity measurement model is established to solve the problem of low accuracy of similarity calculation due to data sparsity.The experimental results show that the accuracy,recall rate and MAE(Mean Absolute Error)have been optimized,which have been improved by 3.4%,5.7%,and 7.8%,respectively.It can be seen that the algorithm can alleviate data sparsity and time dynamic problems,and improve the quality of recommendations.(2)In order to solve the problem of overfitting caused by excessively sparse rating matrices in matrix decomposition,a matrix factorization recommendation algorithm Fusing Tags and Time Information(TTMF)that fuses labels and time information is proposed to enrich a single data source,Alleviate the problem of overfitting in matrix factorization.Firstly,the user's tag preference value and item tag relevance are defined through rating data and tag information,which respectively represent the user's interest in tags,the relationship between tag information and items,and time information is added to indicate changes in user interest over time.The user-item,user-tag,and item-tag matrix models are matrix-decomposed by gradient descent to complete recommendations.The experimental results show that the Rot Mean Square Error(RMSE)of the TTMF algorithm is 7% lower than that of the traditional method LFM(Latent Factor Model).It can be seen that the TTMF model has higher recommendation performance and accuracy.