Research On TopN Recommendation Algorithm Based On Graph Neural Network

With the rapid development of Internet technology,the popularization of cloud computing,big data,the Internet of Things and mobile terminal devices,various data on the Internet emerge one after another,which has led to the explosive growth of data scale.There are more and more channels for ordinary users to receive information,passively entering the era of content overload and data noise.While enjoying the convenience of information acquisition,people are also troubled by information pollution.Based on this situation,the recommendation system came into being.At present,the most widely used recommendation algorithm is the collaborative filtering algorithm.Although the collaborative filtering algorithm can show good performance in most scenarios,it still suffers from the challenges of data sparsity and cold start.As an important research branch of deep learning,graph neural network has developed rapidly in recent years and has become a research hotspot in recent years.This paper mainly uses the graph neural network to obtain more hidden information of users,effectively reducing the cold start and data sparse problems of the recommendation system.The specific research contents of this paper are as follows:(1)Firstly,aiming at the problems of data sparseness and cold start of the traditional collaborative filtering algorithm,knowledge graph and convolutional neural network are introduced to obtain the high-order relationship between users and items in the knowledge graph,so as to reduce the data sparseness and improve the accuracy of the recommendation system.Items and their attributes can be mapped into a knowledge graph to enrich the interrelationships between items.In addition,user-to-user interactions can also be integrated into the knowledge graph,which enables the relationships between users and items,as well as user preferences,to be captured more accurately.Specifically,the model adopts a heterogeneous propagation strategy to explicitly encode the two kinds of information,and applies a convolutional neural network to distinguish the weights of different knowledge graph-based neighbors.Experiments show that this model has obvious improvement compared with the traditional collaborative filtering algorithm.Compared with the best benchmark model,AUC is improved by 3.1%,1.8%and 1.0% on the three datasets of Book-Crossing,Movie Lens-20 M and Last.FM respectively;F1 is improved by 2.7%,1.8% and 2.2% respectively.(2)In order to further mine the potential interest preferences of users,the multi-behavior data of users is introduced.By using the graph structure to represent the user's multi-behavior data,make full use of the user's different behavior data.Capture the different effects of different behaviors on the user's target behavior.This can alleviate the problem of user behavior sparsity.At the same time,an attention mechanism is introduced to distinguish the weight of different behaviors on the target behavior,thereby improving the performance of the recommendation system.Experiments show that using the multibehavior data of users does have a significant improvement over the model using single-behavior data.Compared with the best benchmarks,the average improvement on the Tmall dataset is 9% and 16%,respectively;on the Beibei dataset,the average improvement is 5% and 2%.