Research And Application Of Time-aware Recommendation Algorithm Based On Text Information

Recently,with the popularity of the Internet,the amount of data generated online has been increasing dramatically.In this context,recommender systems have become an effective way to solve the problem of "information overload" because of its ability of active screening information,and have been widely used in various fields.However,the existing recommender systems still face many challenges,such as cold start,data sparsity,and algorithm scalability.As the core of recommender systems,the performance of recommendation algorithm directly affects the quality of the whole system.Aiming at the problems of cold start and data sparsity in existing recommender systems,this paper adopts the methods based on deep learning technology and attention mechanism to solve these problems and improve the accuracy of recommendation.The main work is reflected in following three aspects:(1)We study a Time-aware Recommendation Algorithm Based on Convolutional Neural Network Regressor(TCNNR).By fusing convolutional neural network and random forest regressor and introducing time information,TCNNR constructs a rating prediction model that can learn text,non-text and time information simultaneously.Experimental results on Movielens datasets show that the accuracy of the TCNNR algorithm are significantly superior to various baseline recommender systems.(2)We study a Time-aware Recommendation Algorithm Based on Dual-Attention Mechanism(DAMTR).DAMTR improves the algorithm by constructing two parallel networks to process user review text and item review text respectively,and adds a two-layer attention mechanism and time factor to the network.The algorithm can extract item features and user preferences from review text,and consider the phenomenon that user preferences decay over time.Experiments are conducted on the Amazon datasets containing review text.Experimental results show that the DAMTR algorithm is indeed effective in rating prediction,and its accuracy are better than many mainstream methods.(3)We design and implement a personalized movie recommender system based on DAMTR algorithm.Starting from the needs of users,we design the system architecture,functional modules and database.Finally,we choose the currently popular framework such as Springboot to implement the system.The implementation of this system not only provides convenience for users to choose movies,but also contributes to the application and improvement of DAMTR recommendation algorithm.