图神经网络在推荐系统中的多场景应用综述

A survey of multi-scenario applications of graph neural network in recommendation system

推荐系统能够快速有效地从纷繁复杂的数据中获取有价值的信息.传统推荐在新用户和新项目方面受限,数据稀缺导致推荐困难,同时忽视用户兴趣演化.鉴于传统推荐方法在面对复杂、大规模和动态变化的推荐场景下存在局限性,图神经网络技术引发了学术界的广泛关注.图神经网络在处理图数据和复杂交互关系方面具有优势,并且能够提高推荐的个性化、可解释性和时效性.图神经网络基于图结构数据和节点之间的交互关系,利用节点的特征和邻居信息进行推荐,提高了推荐的准确性.首先,介绍了图神经网络的基本原理和常用模型.其次,分析当前图神经网络在推荐中面临的问题,探讨应对方法.最后,按照不同推荐场景进行划分,对现有研究中的应对方法进行分类归纳.

The recommendation system can quickly and effectively obtain valuable information from complex data.Traditional recommendation is limited in terms of new users and new projects,and data scarcity makes it difficult to recommend,while ignoring the evolution of user interest.In view of the limitations of traditional recommendation methods in the face of complex,large-scale and dynamic recommendation scenarios,graph neural network technology has attracted wide attention in academia.Graph neural networks have advantages in dealing with graph data and complex interactions,and can improve the personalization,interpretability and timeliness of recommendations.Based on the interaction between graph structure data and nodes,graph neural network uses the characteristics of nodes and neighbor information to recommend,which improves the accuracy of recommendation.Firstly,the basic principle and common models of graph neural network are introduced.Secondly,the problems faced by the current graph neural network in recommendation are analyzed,and the countermeasures are discussed.Finally,according to different recommendation scenarios,the coping methods in the existing research are classified and summarized.