A Research Of Recommendation Algorithm Based On Temporal Information

Recently,the electronic commerce has gradually grown into an increasingly mature model.Consequently,information overload is getting worse.Recommender system is the most effective solution to the problem of information overload.The recommender system analyzes users interests based on users historical data,background knowledge,personal experiences and the consumption habits in the system so as to provide the personalized recommendation to the user.Considering the personalized recommendation algorithm,researchers have proposed many methods based on user preferences from different perspectives.However,traditional recommendation methods mainly focus on static recommendation,such as contentbased and matrix factorization.It models user history behavior records in the system as static.Actually,the users preference is evolving over time,so modeling users preference as static will be limited,existing works which consider temporal information in the process of capturing user dynamic interest are usually associated with data sparseness and affecting the effects of recommendation,etc.To solve the above problems,this paper will use the temporal information contained in the recommender system to propose the temporal recommendation algorithm ATCCF which is based on time correlation and the next basket recommendation algorithm LAGCN which is the application of temporal recommendation,respectively.The two models proposed in this paper are compared with traditional recommendation model to verify the correctness and accuracy of the two algorithms proposed in this paper.The main innovations are listed below:First,in the ATC-CF algorithm,we put forward the hypothesis of time correlation based on the influence of internal and external factors on users preferences.We jointly described user preference from the perspective of internal preference and external preference and proposed an enhanced matrix which is effective for weaken the data sparseness in the model and utilized matrix factorization technique to analyze the user preference.Second,in the LAGCN algorithm,which is designed for the next basket recommendation problem,we constructed an item correlation graph to learn correlation between items,and utilized the topology structure among items and global properties of item to learn item embedding in order to learn a robust basket representation.Besides,with the help of user general preference,we extended attention mechanism in the user evolving preference learning process to avoid the noises in the user basket sequence.