Research And Application Of Collaborative Filtering Recommendation Algorithm Based On Graph Neural Network

With the continuous development of information technology,distributed,cloud computing,edge computing and other technologies are becoming more and more mature,giving birth to more and more data.Massive data makes data management and value mining more complex and difficult.With the increasing amount of data,from medical treatment,shopping and entertainment in daily life to the healthy development of the global economy and the defeat of terrorism,we need to rely on the quality of prediction.This requires a system that can predict quickly and accurately,and the recommendation system came into being.Since the recommendation algorithm was proposed,Because of its broad application prospects in different application scenarios,it has attracted many scholars to continuously improve and develop.Therefore,many excellent recommendation algorithms have been born,and very good results have been achieved in practical applications.The main work of this thesis is divided into three parts:(1)The first is to introduce the collaborative filtering recommendation algorithm based on graph neural in the collaborative filtering recommendation algorithm.It aims to solve the problem of long-tail items and data sparseness in traditional recommendation algorithms.By stacking multi-layer embedding propagation layers to fully capture the collaborative signals in high-order connectivity,we can better carry out collaborative filtering recommendation according to the user's behavior of the project and enhance the interpretability.In addition,because the user's different ratings for different items represent the user's different preferences for different items.Therefore,this thesis introduces the attention mechanism in the propagation layer of the Light GCN model and proposes the Light GCN-Att model.By adding a variable weight to better reflect the user's preference for different items,it also increase the interpretability of the model.Secondly,this thesis uses the NCF neural network model to replace the original inner product operation in the prediction layer of the Light GCN model,so that the model can capture the complex nonlinear interaction between users and items,thereby further improving the accuracy of the model.(2)In the experiments,this thesis validates the above models on the Gowalla,Amazon-Book and Yelp2018 datasets.The experimental results show that the Light GCN-Att model learns them by using Light GCN's linear propagation of user and item embeddings on the user-item interaction graph.The weighted sum of the embedding learned on all layers is used as the final embedding.This simple,linear,and compact model is easier to implement and train than NGCF.At the same time,the changes of the propagation layer and the prediction layer can further improve the recommendation effect.(3)Finally,the proposed recommendation model is applied to the gym item recommendation.The recommendation system uses explicit and implicit feedback data at the same time,and integrates the advantages of content-based recommendation,search engine and Light GCN-Att model.It can well solve the problem of data sparsity and the problem of difficult recommendation of long-tail projects.According to the users' different fitness goals and fitness intensity,complete the personalized fitness project recommendation for the target users.