Session Recommendation Algorithm Based On Graph Neural Network Research And Implementation

The recommendation system system can extract products that may be of interest to users from their information and behavior sequences,thereby generating personalized recommendation lists for users.When a user visits a website,the recommendation system server will record the user’s access records during the current session cycle,and finally generate a personalized recommendation list based on the user’s behavior sequence,which is called a session based recommendation system.In recent years,session recommendation models based on graph neural networks have received increasing attention from scholars.Although they can effectively improve the accuracy of recommendation results,existing recommendation algorithms still have the following two problems:(1)Graph neural networks are difficult to accurately capture the long-distance dependencies between nodes,and the self attention mechanism has learned new embedded representations of all items,But it will allocate attention weights to many untrustworthy outputs,which affects recommendation performance.(2)The existing session recommendation methods usually only consider using the current session information,ignoring the association with other sessions,which lacks the global perspective of the target session.In addition,the correlation weights between projects are pre specified based on the edges in the graph and do not participate in the model training process,which will result in the inability to capture the dynamic correlation between projects.In response to the above two issues,this article has designed and implemented two graph neural based session recommendation algorithms:(1)A session recommendation algorithm that combines dual sparse attention mechanism and graph neural network.It identifies irrelevant objects in the current session sequence by constructing a dual sparse attention mechanism,while assigning higher weights to important items.It automatically removes irrelevant sequence objects in the calculation,effectively capturing users’ real interest preferences.(2)An enhanced graph neural recommendation algorithm based on similar sessions calculates similarity by using the number of duplicates between different sessions to construct a session set,and calculates feature interactions between different neighbors inthe global session graph through element level multiplication.At the same time,learnable weight parameters are assigned to the items in the current session,and their weights are continuously optimized during the model training process,thereby achieving dynamic correlation between items.This article not only verifies the reliability and practicality of the proposed model on the Diginetica dataset and the 30 music dataset,but also trains and fine tunes the music domain dataset based on the above model,and builds a commercially available music recommendation system in the real world.