Research On Session-based Recommendation Algorithm Using Deep Learning

With the rapid development of the Internet,the massive amount of information brings convenience to people,but also causes the problem of "information overload".The personalized recommendation system is considered to be one of the effective tools to solve the problem of information overload.Recommendation algorithm,as the core of the recommendation system,has a decisive influence on the recommendation effect.In many online recommendation systems,the interaction between the user and the system is organized into a conversation,which refers to a sequence of interactions between people and the system that occurs within a certain period of time.The conversation-based recommendation algorithm is to predict the user's next action based on the user's current interaction sequence.The combination of session-based recommendation algorithm and deep learning technology has achieved a series of results,but there are also some shortcomings.For example,the exploration of deep learning models into contextual information is not deep enough;for example,when using cyclic neural networks,cyclic neural network memory The capacity is limited and the memory access is not flexible enough.These problems affect the performance of the recommendation algorithm to a certain extent.This article researches and explores the above problems and proposes some improvement measures.The main work is as follows:(1)A context-aware recommendation algorithm that integrates user session data is proposed.The context information is mapped into low-dimensional real vector features through embedding,and the low-dimensional vector features are integrated into the session-based session through three combinations of Add,Stack,and MLP.The recurrent neural network recommendation model is designed to dynamically describe the user preferences in the conversation sequence based on the BPR loss function to improve the ability of personalized recommendation.The design experiment explores and compares the influence of different combinations of contextual information on the recommendation effect.The effect of integrating contextual information into the model at different stages of the GRU unit on the recommendation performance is studied.The experimental results show that the context-aware recommendation algorithm fused with user session data proposed in this paper has a significant improvement in recall rate and MRR on the public data set compared to the baseline algorithm GRU4 REC and other algorithms.(2)Aiming at the problems of limited memory capacity of cyclic neural network and insufficient memory access flexibility,a conversation recommendation algorithm combined with memory network is proposed.The algorithm designs a hierarchical recommendation model.The model is divided into two layers,which are the session-level GRU model that characterizes the current conversational interest and the user-level memory network model that characterizes the user's long-term interest.At the same time,for the flexible access of the memory vector in the memory network,the attention mechanism is introduced,and the corresponding writing and reading modules are designed.Experimental results show that the model can better capture the long-term interests of users.Experiments on the public data set prove that in the performance improvement comparison of the number of sessions of 10 versus the number of sessions of 5,the proposed algorithm HNUM in this paper has 4 improvements in Recall and MRR compared to the baseline algorithm HGRU.