Research And Application Of Recommendation Algorithm Based On Graph Convolutional Neural Network

The emergence of the big data era provides both opportunities and challenges for various aspects of work and life.While it offers a broad perspective,it also leads to information overload.To address this issue,recommendation algorithms have been widely adopted in shopping platforms,social media,advertising,and online search fields.These algorithms can help users save time and improve efficiency by reducing search and screening time,enhancing purchase conversion rates,improving user stickiness,increasing platform activity,and boosting user satisfaction.Traditional recommendation algorithms represent users and items as low-dimensional vectors,which may lose high-dimensional interactive information in graph data.Recommendation algorithms based on graph convolutional neural networks can directly learn graph-structured data,thus improving the learning of high-dimensional information in graph data.Considering this characteristic of the graph convolutional network,this thesis applies an improved graph convolutional network to the recommendation algorithm and makes a practical application of the improved graph convolutional recommendation algorithm.The main contributions are as follows:(1)A recommendation algorithm based on a graph convolutional neural network is proposed.The improved graph convolution method is used to make the graph convolutional network more suitable for collaborative filtering recommendation tasks.To address the problem that the existing graph convolutional network cannot flexibly handle graph datasets of different scales,a normalization parameter is introduced.This parameter controls the influence weight ratio of neighbor nodes and its own nodes during neighborhood aggregation.Finally,experiments on three datasets of different scales(Gowalla,Yelp2018,and Amazon-book)verify the effectiveness of the algorithm on graph datasets of different scales.The value of the normalization parameter is also discussed through experiments.(2)A graph convolution recommendation model based on dynamic negative sampling is proposed.The existing graph convolution recommendation algorithm lacks an effective negative sampling strategy.This makes it difficult to use negative samples to improve the model.To solve this problem,a dynamic negative sampling strategy is adopted to randomly select a negative sample set from the item nodes that the user has not interacted with.The score of the negative sample is obtained through graph convolution,and hard negative samples are selected by comparing the sample scores in the negative sample set.Bayesian personality ranking is used to calculate the difference between positive and negative samples as a loss for model optimization.Experiments verify that the negative sampling strategy can improve the recommendation results of the graph convolutional network.(3)The design and implementation of a product recommendation system based on the graph convolution algorithm are completed.Firstly,the requirements of the system are analyzed,and the functional and non-functional requirements of the system are clarified.Then,the general design is carried out,including the overall architecture of the system,business process,and database design.Then,the recommendation module of the system is introduced in detail,and the actual application system interface and business process are shown.The practical application value of the proposed algorithm is demonstrated by implementing the product recommendation system.