Design And Implementation Of Conversational Recommender System Based On Graph Embedding And Attention Mechanism

At present,conversational recommendation system based on graph neural network is a new research direction in the field of recommendation system.The current research focus on recommendation systems is how to use as little user information as possible to generate accurate and efficient recommendation results,and conversational recommendation is very suitable for the current Internet status.A session is based on a user’s historical behavior sequence within a certain period of time.The main task of the conversational recommendation system is to use the conversational sequence to predict the next item that the user may interact with.The current session recommendation algorithm based on graph neural network only considers a single user session graph to generate item feature vectors,and does not consider the conversion relationship between items in other session sequences,and the amount of extracted information is limited.In the process of using item feature vectors to model user interests,it is impossible to model and represent user interests more accurately.Based on the above problems,this paper carried out related innovations and experiments.The specific research content and research results are as follows:(1)A session recommendation algorithm based on temporal and global session information enhancement is proposed.Compared with existing session recommendation algorithms,this method extracts crosssession item transformation relations by constructing a cross-session global session graph,thereby providing richer contextual information for prediction.Aiming at the defect of adjacency matrix,a recurrent neural network is proposed to preserve the sequential information of sessions.In the process of user interest modeling,a global conversational interest is proposed,and the user’s interest is modeled from the perspective of the global conversation,so as to improve the quality of the user’s interest representation,and then improve the effect of the model.(2)A conversational recommendation algorithm based on contrastive learning and category feature information enhancement is proposed.The algorithm mainly uses the concept of data augmentation of contrastive learning to augment the original training data and construct a "negative sample" session sequence to improve the model’s discrimination ability.In the process of user interest modeling,an item category feature extractor is designed to extract the item category information in the user session,so as to model the user’s item category interest preference.(3)Design and implement a personal movie blogging system based on session recommendation algorithm.The system develops a front-end and back-end interface based on the Web,and provides users with the functions of registration and login,blog posting,blog evaluation and movie recommendation.The system integrates the models trained in this paper to generate movie candidate sets personalized for users.