Research On Sequence Recommendation Based On Global Enhanced Graph Neural Network

Most of the existing session based recommendation systems recommend based on the correlation between the last clicked item and the user preference of the current session,and ignore that there may be item transitions related to the current session in other sessions while these item transitions may also have a certain impact on users’ current preferences;hence,it is indispensable to analyze users’ preferences comprehensively from the perspective of local session and global session.Furthermore,most of these recommendation systems ignore the importance of location information,whereas items closer to the predicted location may be more relevant to the current user’s interests.To solve these problems,this paper proposes a recommendation model based on global enhanced graph neural network with LSTM(GEL-GNN,global enhanced graph neural network with LSTM).GEL-GNN predicts the behavior of users according to all sessions.The GNN is employed to capture the global and local relationship of the current session while the LSTM is employed to capture the relationship between sessions at the global level.Initially,users’ preferences are to be translated as a combination of conversation interests based on global and local levels through the attention mechanism layer.Then,the distance between the current position and the predicted position is measured with the reverse position information,so that the user behavior can be predicted more accurately.A number of experiments are conducted on three real data sets.The experimental results show that GEL-GNN is superior to the existing session-based graph neural network recommendation models.Finally,based on the GEL-GNN model,a personalized book recommendation system is designed and implemented.The front end uses the Vue framework,the database uses My SQL,and the back end uses Python’s Django framework.The designed and implemented personalized book recommendation system can recommend books that may be of interest to users according to their current and long-term preferences.