Article Recommendation System Based On Text Processing And User Profile

With the advent of data information age,an endless number of article platforms and websites make it more convenient for people to acquire knowledge.But at the same time,the rapid expansion of information also brings some problems.On the one hand,it is difficult for users to find targets quickly in the face of massive articles;On the other hand,the platform also wants to present personalized content to the user to enhance the user experience.As a way of information filtering,article recommendation system has greatly improved the efficiency of reaching users and targets,and has gradually become an indispensable part of Internet products.Among them,text and user,as the two most important object entities,always affect the final recommendation effect.Therefore,it is of great significance to study the recommendation method based on text processing and user profile.Firstly,considering the semantic richness of the text in the article recommendation,this thesis proposes a recommendation model based on deep text processing by comparing with various existing text processing methods.In view of the problem that the existing methods seldom consider global word co-occurrence,the model uses GCN(Graph Convolutional Network)network for text classification and extract the feature.In addition,according to the dynamic and relevance of interest preference which is a feature dimension of user profile,the user interest model is constructed and improves the accuracy of feature extraction and optimizes the final recommendation effect.Then,considering that traditional collaborative filtering methods usually have problems such as cold start of users and too sparse matrix,this paper designs a collaborative filtering recommendation method that integrates user profile technology after comparing and investigating various optimization methods applied in article recommendation scenarios.In this method,a user profile labeling system is designed for the recommended scenes,and multi-dimensional user features are extracted by using relevant text processing technologies.An optimized user clustering method is proposed,which can effectively classify users and greatly improve the efficiency of recommendation computing.Finally,the problems of cold start and matrix sparsity are effectively alleviated when collaborative filtering recommendation is carried out.Finally,the simulation experiments of the two methods are carried out,and the proposed methods are implanted into the actual recommendation system with a parallel relationship.Then,the final recommendation is completed by mixing and sorting the two kinds of articles to be recommended.The experimental results and the actual operation of the system all prove the feasibility and accuracy of the method.