Research And Implementation Of Collaborative Filtering Recommendation Algorithm Based On Semi-Supervised Learning

With the rapid development of information technology.In the era of mass information,how to effectively deal with the information overload caused by inaccurate information has become one of the most popular research topics in this era,collaborative filtering recommendation algorithm can alleviate this problem to a certain extent.Therefore,this method has been widely used in many fields in recent years.But in most cases,the interaction matrix between users and projects is sparse,there is a lot of unlabeled data and often accompanied by data imbalance,which has a great impact on the recommendation effect.The current optimization schemes mainly focus on the improvement of similarity algorithm,matrix dimensionality reduction,missing score prediction and hybrid recommendation methods,the missing score prediction is to make up the missing value of the scoring matrix by the method of prediction,so as to reduce the sparsity of the matrix.However,this method needs to predict each missing value,which results in high time complexity.In order to solve the above problems,the following work is done in this paper:Firstly,in order to improve the computing efficiency and data sparsity,this paper proposes a user ranking recommendation framework based on collaborative filtering,which uses user interaction data and interaction time to partition users.An improved similarity calculation method is adopted for users with rich and reliable interactive data,which can improve the accuracy of recommendation and reduce the computational load of the system.The collaborative recommendation method based on semisupervised learning proposed by users with sparse data can alleviate the problem of data sparsity.onsidering the two-way interaction between users,it is added into the similarity calculation,and the method of similarity calculation is classified dynamically,make positive or negative adjustments to the similarity between users by using the common user ratings.The experimental results show that the similarity calculation method can improve the overall recommendation and TOP-N recommendation to a certain extent.Then semi-supervised learning is used to supplement the sparse data,and a collaborative recommendation method based on semi-supervised learning is proposed.This method firstly uses the FCM method to cluster,and selects the samples near the hyperplane as the input of Lap-SVM,the accuracy of clustering results is improved and the complexity of Lap-SVM algorithm is reduced.Then,Lap-SVM algorithm is used to train the model,and a posteriori probability knowledge is added to mark the location of samples,the classification deviation of fuzzy anomaly samples is reduced,and the sample credibility weight is provided for matrix decomposition.Finally,the classification result of unlabeled samples is added to user-item matrix,the predicted samples were weighted with a posteriori probability label,the matrix was decomposed with ALS,and the unknown items in the item-score matrix were predicted,the collaborative recommendation method based on semisupervised learning has good performance in many evaluation indexes.At last,this paper develops and designs the collaborative filtering video recommender system based on semi-supervised learning,expounds the functional requirements of each module of the system,and applies the method proposed in chapter 4 and Chapter 5 to the video recommender system,the function of each module of the system is tested.The test results show that the recommender system has a certain degree of reliability.At last,the main function pages of the system are shown.