Research On Personalized Recommendation Algorithm Based On Combining User's Interest With Time Effect And Its Application

In recent years,many scholars have proposed to alleviate the problem of information overload by building a personalized recommendation system to recommend effective information to users.High-performance recommendation algorithms are the core of recommendation systems.At present,the recommendation algorithms mainly faces the problems of data sparsity,cold start,interest drift and time effect.To solve the above problems,this thesis will study the user's interest and time effect based on the CF algorithm and LSTM algorithm,respectively.The thesis is constructed as follows.We first propose a personalized recommendation algorithm(IT-CF)based on the CF algorithm to study users' potential interest and time effect.This algorithm uses the tag information to mine the users' potential interests,calculates the users' preference weights of item tags by using TF-IDF,and combines the Ebbinghaus forgetting curve with the time window function to obtain the temporal weight function.The UserCF algorithm can be improved by using new tag preference weights and new temporal weight functions,but the Item CF algorithm is improved by using new temporal weight functions only.Finally,the prediction rating of the two improved algorithms are mixed to obtain the final recommendation list.By comparing with the UserCF algorithm,Item CF algorithm and two algorithms with similar research directions,the experimental results show that the IT-CF algorithm improves the accuracy of the prediction results and optimizes the quality of the recommendation results.This thesis also proposes another personalized recommendation algorithm(RT-LSTM)based on the LSTM algorithm to study user ratings and user behavior time interval information.This algorithm designs a T-LSTM model to analyze the effect of time interval information of user behavior on user interest by adding a time gate on the LSTM.Thus we uses the T-LSTM to build the user's sequential behavior model and uses the ratings as the determining criterion of the user's preference for items,and combines the ratings with the prediction results of the next moment to ensure that the predicted items are preferred by the user.Through the analysis of experimental results,the short-term prediction success rate of the RT-LSTM algorithm compared with similar algorithms is improved.Next,this thesis builds a personalized video recommendation system based on the distributed system architecture.First,we explain the requirements of the system from the user's perspective,construct the structure of the recommendation system,design a recommendation engine based on the above two algorithms,integrate the recommendation engine into the system.Finally,we establish the database and complete the development of function modules.