Research On Recommendation Algorithm Based On Deep Learning And Time Context

Information recommendation technology is an important technological measure in modern informatization society.More useful information is provided by analyzing users' preference and characteristics of the product,which makes the experience of website and client-side application perform much better,thus reducing unnecessary information delivered,solving the problem of information overload,improving the efficiency of information transmission and information-interest commercial conversion rate.The module of recommendation algorithm is the key to recommendation system,which decides upon the merits and demerits of a recommendation system.As a classic recommendation algorithm,collaborative filtering algorithm is widely used,which can be divided into user-based collaborative filtering(UBCF)recommendation algorithm and item-based collaborative filtering(IBCF)recommendation algorithm.However,due to the following problems,the accuracy of the recommendation result is affected,which causes too large amount of data leads to too sparse user behavior data and difficulty in extracting data features;Considering a single data perspective,only user/project-score information is considered.In response to the above issues,the following work has been performed:(1)For the data sparsity caused by too large data volume and little user behavior,in this paper,it proposes UDBN-UBCF recommendation algorithm witch combined unsupervised deep belief network with item-based collaborative filtering and UDBNIBCF recommendation algorithm witch combined unsupervised deep belief network with user-based collaborative filtering.Using the input user-score/item-score highdimensional data,the UDBN model is used to reduce the data dimension,and the lowdimensional data output by the model is used for similarity calculation,and then scored,and the results are recommended.Using Movielens datasets for comparison experiments,we can see from the experimental results that UDBN-IBCF algorithm and UDBN-UBCF algorithm can effectively alleviate the problem of excessive recommendation error caused by data sparse compared with IBCF algorithm and UBCF algorithm.(2)Regarding the data that only considers the limitation of the user's rating of the item,the time context information is added,and an IR-IBCF algorithm is proposed that combines item-related forgetting curves.The IR-IBCF algorithm adds the itemdependent forgetting function based on the Ebbinghaus Forgetting Curve to the traditional IBCF algorithm.It pays more attention to the time correlation between items and improves the prediction accuracy of the original algorithm.(3)According to the UDBN-IBCF algorithm mentioned above,time context information is added to further improve the UDBN-IR-IBCF algorithm that combines item-related forgetting functions.Through experimental simulation,comparing the error values of the two improved algorithms and the original algorithm in the same data set,it can be shown that the UDBN-IR-IBCF algorithm can solve the problem of data sparseness and single attribute at the same time,further improving the accuracy of the algorithm.