Research And Application Of Session Sequence Recommendation Technology Based On Deep Learning

Recommendation system plays a very important role in today’s data explosion society.It can help people select the most concerned information from the complex data and help users make the most appropriate decisions and choices.The emergence and development of recommender systems have brought great benefits to people,not only improving people’s utilization of information,but also improving the quality of services.At present,many models and technologies of recommender systems have been proposed and have achieved good results.Session-based recommendation system,as a sub-field of recommendation system,aims to make real-time recommendation to users,and generate a list of items to be clicked next time for users according to the click sequence of items within a short period of time.However,in the actual scene,the user’s click sequence behavior is very complex,so when modeling the sequence behavior,it is difficult for most models to capture the complex item transfer relationship;secondly,the existing session-based recommendation In the model,the context information of the item is insufficiently utilized,so that the potential information cannot be fully obtained during the recommendation,resulting in inaccurate recommendation results.In response to the above problems,this paper proposes two session-based recommendation models to improve the above problems.(1)This paper presents a conversational recommendation model(SR-GGNN)that combines GRU and graph neural network.This model first constructs a directed graph of the session sequence,and then gates the graph neural network to model the session,so as to obtain the complex relationship between the item sequences clicked in the session;then the obtained item vector is input into the GRU network,It is used to extract the sequence information of the session items,and the last click of the output is used as the user’s short-term interest;then the vector output by the GRU unit is used to capture the global preference of the session sequence through a layer of attention mechanism,and finally the current user’s global preference The preference and short-term interest are combined to achieve the final recommendation result.(2)A conversational recommendation model(SR-AGGNN)integrated with auxiliary layer graph neural network is presented.This model is further improved on the basis of SR-GGNN.While modeling the sequence of items in the session,the type sequence of the item is modeled as auxiliary layer information and fused into the vector of the item as a new vector;then the fused vector is input into the GRU network,and the output is The last click is taken as the user’s short-term interest;then the vector output by the GRU unit is used to capture the global preference of the session sequence through a layer of attention mechanism,and finally the current user’s global preference and short-term interest are combined to achieve the final recommendation result.(3)Finally,experiments are conducted on two real public Yoochoose datasets and Diginetica datasets,comparing both models with baseline methods.Compared with other benchmark models,the experimental results show that the two model methods in this paper have better effects,and the overall evaluation index has been improved,which also proves the effectiveness of the model in this paper.