A Matrix Factorization Recommendation Model Incorporating The Emotional Intensity And Temporal Characteristics Of Social Tags

With the popularization of social networks and multimedia content sharing websites,information overload and data expansion have greatly restricted the development of user experience.The collaborative filtering recommendation system realizes personalized recommendation by analyzing the user's historical behavior data,but the traditional collaborative filtering algorithm relies too much on the user's historical rating data,and the cold start and data sparsity problems are prominent.As an important user behavior,social labeling is the core and foundation of the recommendation of the social labeling system,and it has the dual characteristics of content and relevance.The matrix factorization technology reduces the dimensionality of the original matrix features and can be effectively integrated with social tags,and has strong scalability and high prediction accuracy.Currently,there are mainly problems in the recommendation of social tagging systems,such as the coupling of complex emotional semantics of social tags,complex interactions between sparse feature inputs,and nonlinear complex interactions between users.Therefore,this paper integrates the research of social recommendation technology,social tag technology,matrix factorization technology and sentiment analysis technology,and proposes a matrix factorization recommendation model that incorporates the emotional strength and timing characteristics of social tags.First,based on the mixed expert model,an sentiment analysis method that can be used to identify and quantify the emotional strength of social tags is proposed which combines random forest,neural network and light gradient boosting machine learning methods to output a user's fine-grained emotional profile matrix.Then the ternary dynamic and complex topological relationship between users,resources and social tags is integrated into the process of matrix factorization.The emotional intensity recognition of social tags and temporal feature mining are used to expand the potential feature factors between the coupling items of the social tag matrix factorization model,and the social regularization technology is used to constrain the social tag matrix factorization process.At the same time,the traditional user similarity measurement method is improved,and the local relationship of users is merged with the global model of social label matrix factorization,and finally the score prediction and resource recommendation function of the matrix factorization recommendation model of social label emotional strength and time series characteristics are realized.Finally,this paper verifies the STSMF model experimentally on the two public data sets of Last.FM and Movie Lens,compares and analyzes the current similar advanced algorithms.We use MAE,RMSE,Recall,Precious and other indicators for performance evaluation and sensitivity analysis.Experimental results show that the social tag sentiment analysis proposed in this paper is effective in mining fine-grained user sentiment,and the recommendation prediction performance of the STSMF model is better than similar advanced methods.