Research On Session Recommendation Based On Graph Neural Network And Attention Mechanism

With the development of the internet industry and the popularization of big data technology,a large amount of information is rapidly spreading in various forms in the network,and the storage volume of data is increasing exponentially.As a result,people have ushered in the era of information flood.The application of information explosion and big data technology has promoted the rapid development of recommendation systems.Recommendation systems can solve the problem of information overload by analyzing users’ preferences and recommending information that interests them.However,in today’s numerous real scenarios,recommendation systems are facing increasingly transparent user information,which is why session based recommendation systems have ushered in a wave of research.The session recommendation system does not rely on user configuration information and only predicts user preferences based on user historical behavior data.In the field of conversation recommendation,graph neural networks are widely used due to their powerful ability to explore graph structure information.However,previous session recommendation methods based on graph neural models had some shortcomings: firstly,they did not fully utilize other session information in modeling project transformation;Secondly,it is not possible to accurately activate users’ core interests from noisy data.This article conducts in-depth research on the above issues and proposes corresponding improvement strategies.The main work and achievements are as follows: Firstly,in response to the insufficient utilization of session sequence information in existing models and the shortcomings in project transformation information,this article proposes a session recommendation model based on local neighborhood graph information and attention mechanism.This model captures user interests and preferences from both the current session and neighboring sessions,providing more auxiliary information for the model through collaborative thinking.The model uses a simplified graph convolutional layer neural network to aggregate node information and capture project transformation information.In the session presentation layer,the influence of target items is considered,and the attention mechanism of target item perception is introduced to optimize the session presentation.Secondly,a session recommendation method combining neighborhood enhancement graph models is proposed to address the interference of noisy data in neighborhood sessions on the representation of model learning items.The model constructs a neighborhood graph based on neighborhood conversations,designs an attention mechanism for neighborhood perception,and strengthens the correlation representation of neighborhood nodes;Then it is fused with the item representation of the current session to get the final item representation.At the presentation layer of the session,target awareness attention is fused to extract target relevance.Thirdly,implement the development of a prototype system based on session recommendation,using deep learning technologies such as B/S architecture and Tensorflow to build and deploy a movie recommendation system.The system can implement similar movie recommendation and recommendation for you through offline training model based on the interactive information between users and movies.