Research On Graph Neural Network Recommendation Algorithm Based On Session

In the context of today’s big data,where billions of levels of data are generated every day,the question of how to extract effective information from the vast amount of data is a highly visible one.With the development of technology,from the very beginning of data mining to the current big data and data analytics,all are aimed at solving this problem,and recommendation systems are the cornerstone of big data.The main task of a recommendation system is to predict the next item that a user is likely to interact with based on the user’s history.Early models,based on collaborative filtering and Markov chains,have achieved some results while also revealing problems in some cases.In recent years there has been a lot of interest in graph-neural based recommendation models,which apply deep learning ideas to the recommendation domain,leading to much better results,but still missing some key information.First,in the domain of session-based recommendation,most recommendation models focus only on the influence of node neighbours and ignore the influence of higher-order items in the current session,which leads to the inability to dig deeper into the hidden information in the session,making the model unable to accurately capture user preferences.Secondly,in the field of session recommendation,existing models are either too simple or introduce noisy data for neighbouring sessions,which makes it impossible to obtain accurate information about items across sessions and limits the model’s ability to capture user preferences,resulting in poor recommendation results.In this thesis,we conduct an in-depth study of current session-based graph neural network recommendation models,and propose a session-based graph location attention neural network recommendation model and a session-based similar neighbourhood graph neural network recommendation model.The main work of this thesis is as follows.1.we study the relevant scientific literature on traditional recommendation models,session-based recommendation models and graph neural network-based recommendation models,and conduct an in-depth study and analysis of the models proposed in the literature,putting forward their problems and possible optimisations.2.A session-based positional attention recommendation model GPAN is proposed,which models high-order and low-order items in the current session separately and explores the hidden item information in the session in a deeper way.A new method for representing the location of items in a session sequence is proposed,which incorporates the frequency of items in a session to more accurately represent the importance of the location and frequency of items in a session,and achieves better recommendation results.3.A session-based similar neighbour graph neural network recommendation model SNGN is proposed,which presents a new way of constructing neighbouring sessions,extracting valid information while reducing noisy data.The fusion of neighboring sessions and current sessions is modeled separately to obtain more accurate user preferences and achieve better recommendation results.