Research On Recommendation Algorithms Based On User Preference And Influence

In the era of information overload,it is very difficult to find the interested information from a large amount of information.As an important tool to alleviate this problem,the recommendation system can mine users' preference information from large-scale data to provide users with accurate personalized recommendation services.Currently,popular recommendation algorithms use user ratings,comments,and other subjective data that clearly reflect user preferences to predict user interests.However,such data is often sparse and will limit the recommendation quality of recommendation algorithms.In addition,how to dynamically capture the changing preferences of users and improve the diversity of recommendations is also an urgent issue in the research of recommendation systems.Based on the problems in existing research work,this paper conducts in-depth research on hidden information in user's historical behavior,explores user interest from multiple perspectives of user history preference and influence,and optimizes the recommendation algorithm.The main research contents are as follows:1.An improved collaborative filtering recommendation algorithm based on user behavior and degree center influenceThe traditional collaborative filtering recommendation algorithm only calculates the similarity based on user-item scoring matrix and generates recommendation results.The recommendation effect is not ideal and the considerations are too single.To solve the above disadvantages,this paper designs an improved collaborative filtering recommendation algorithm.Based on traditional collaborative filtering algorithm,a time function that adapts to changes in user preferences and influence factors of network users is introduced.The time weight-based scoring matrix is integrated with the influence matrix to alleviate the sparseness problem of scoring items.It has better recommendation quality than traditional collaborative filtering algorithms.2.A bidirectional GRU neural network recommendation model based on attention mechanismTraditional methods face data sparseness and cold start problems when predicting user preferences.Considering that deep learning technology can effectively capture the nonlinear relationship between users and items.A recurrent neural network model based on attention mechanism is designed to deal with the long-term dependence problem in historical behavior of users,and model user's long-term and short-term preferences.First,the self-attention mechanism is used to learn user's short-term interaction trajectory and distinguish the importance of each potential feature or factor;Then,based on bidirectional GRU neural network to learn long-term stable preferences of users,and uses attention mechanism to activate local preferences related to target item,so as to model user preferences changes;Finally,the long-term and short-term preferences are combined to predict user's next behavior.This model can effectively model user's long-term and short-term preferences and capture user preferences more accurately,it shows better performance on different evaluation indicators.3.A recommendation model based on user long and short-term preference and user influenceUsing degree centrality in complex network to describe user influence and applying it to recommendation can alleviate cold start problem in datasets to some extent,but the measurement method is too simple and the improvement of recommendation effect is not obvious.Therefore,the paper proposes an improved network model combining local centrality influence measurement method with coefficients,which combines influence factors with user short-term and long-term preference.This paper uses Transformer to replace bidirectional GRU neural network to mine user's long-term and short-term interest preferences implicit in implicit characteristics of user historical interaction.Data analysis and experiments show that the model is superior to compared model in both training efficiency and recommendation accuracy,thus verifying the effectiveness of combining users' long-term and short-term preferences and user influence to make recommendations.