Personalized Movie Recommendation Model Based On Text Sentiment Analysis

With the rapid development of e-commerce,the recommendation system has become an indispensable tool.At present,the main problem of the recommendation model are the strong sparsity and high dimension of the original data,which lead to low personalization and low computational efficiency.Traditional algorithm directly utilizes the behavioral data that users have seen watched,but audiences score freely with their moods and some producers raise their earnings by scoring maliciously.These behaviors affect the accuracy of the recommendation results.In view of the above problems,this thesis studies the most mainstream collaborative filtering algorithm,and introduces text sentiment analysis to improve recommendation accuracy.The main research contents include:1.For the problems of strong sparse and high dimensional,matrix decomposition is introduced based on the K nearest neighbor model to form a hybrid algorithm.First,the Pearson similarity is improved to take into account more implicit features,which can more accurately calculate the respective "neighbors" collection of users and projects.Then use the two sets of "neighbors" to construct the scoring matrix.This process filtering the information in advance to avoid traversing other sparse data,thus reducing the impact of sparse data on the experiment.The matrix decomposition method can speed up the calculation,reduce the time complexity and improve the recommendation accuracy.2.The traditional recommendation algorithm only analyzes historical user ratings,and the quality of these data affect the accuracy of the recommendation.Filter the pan-push results through film-based sentiment analysis to achieve accurate recommendations.Aiming at the context information of the film review with its own attribute characteristics and word order irrationality,a feature-enhanced deep learning model is introduced.A weight distribution layer is introduced between the input layer and the convolution layer to analyze important parts,reduce noise,and improve processing characteristics.The convolutional layer is used to extract different local features,and a gating mechanism is added to the convolutional layer to reduce the risk of gradient dispersion caused by the gradient descent method.Finally,the sequence annotation layer is added to enable high-level abstract features to represent correct text semantics.Experiments show that the character and word granular vector are combined into input,which solves the problem of ambiguous word segmentation,and proves that the use of this model for film evaluation sentiment analysis is better than other models.Finally,the results of the text sentiment analysis are combined with the recommendation model to enhance the robustness of the model and improve the accuracy of personalized recommendation.