Research On Collaborative Filtering Algorithm Based On Chinese Comment Mining

With the advent of the big data era,recommendation systems play an important role in solving information overload.Collaborative filtering,as a recommendation method that utilizes massive user ratings for collaborative processing,has been widely applied in various fields.User reviews not only contain item attribute information,but also depict user preferences for items from multiple perspectives.Therefore,mining aspect information from user reviews has become an important means to improve recommendation system performance.Aspect information in reviews is fine-grained and can reflect user interactions and preferences for items from multiple perspectives.Most current review-based recommendation systems tend to focus on linearly learning the contribution of review information to recommendations,neglecting nonlinear learning of multi-aspect information.Moreover,due to the significant differences between Chinese and English rules,it is difficult to directly apply English-oriented models to Chinese-oriented paradigms,and there are fewer recommendation models based on Chinese review aspects.In response to the above background and issues,this study investigates aspect-level recommendation algorithms based on Chinese movie reviews and proposes two collaborative filtering methods based on mining aspect information from reviews.The methods are validated and analyzed on a publicly available Douban movie review dataset.The main innovations of this thesis are as follows:In order to utilize aspect sentiment information to improve the recommendation accuracy of neighborhood-based collaborative filtering,this thesis proposes a user collaborative filtering model based on Chinese movie review aspect sentiment analysis.To improve the accuracy of aspect sentiment analysis in Chinese movie reviews,a rule-based aspect sentiment analysis method is first proposed.This method extracts aspect terms based on Chinese dependency syntax and the frequency of aspect words in movie reviews.Then,by manually annotating a small dataset of Chinese movie aspect sentiment,a pre-trained model is fine-tuned for accurate sentiment polarity classification.Then,the proposed aspect sentiment analysis method is combined with user collaborative filtering,reducing dimensions based on aspects and introducing user preferences for aspects to predict recommendation scores.Experiment results show that this model outperforms the baseline model in recommendation accuracy.In order to utilize aspect interaction information to improve the recommendation accuracy of deep learning-based collaborative filtering,neural collaborative filtering models based on aspect implicit features and explicit features perception are proposed.First,a neural network framework based on generalized tensor factorization is implemented,using aspect term information to construct a user-item-aspect interaction three-dimensional tensor.On this basis,the tensorized multilayer perceptron is combined with generalized tensor factorization,introducing aspect implicit features into neural collaborative filtering,capturing user-itemaspect interactions in a mixed linear and nonlinear manner,and implementing a tensor factorization technique generalized to the aspect-based recommendation domain.Second,aspect explicit features are introduced into neural collaborative filtering,constructing a useritem-aspect sentiment three-dimensional tensor,and implementing a neural collaborative filtering model based on aspect sentiment polarity.In general,by achieving a two-layer focus on aspects information and global focus on three-dimensional data,the technology is integrated into the aspect-based recommendation domain,to some extent solving the cold-start problem of collaboration.Experiment results show that both models outperform the baseline model in recommendation accuracy and effectiveness.