多阶元路径引导的异质图神经网络新闻推荐模型  

Meta-path guided heterogeneous graph neural networks for news recommendation

作　　者：王菲菲 林中潭 吴昆 韩树庭 孙立博 吕晓玲[1,2] WANG Feifei;LIN Zhongtan;WU Kun;HAN Shuting;SUN Libo;LÜXiaoling(Center of Applied Statistics,Renmin University of China,Beijing 100872,China;School of Statistics,Renmin University of China,Beijing 100872,China;Recommendation Group,Bytedance,Beijing 100024,China)

机构地区：[1]中国人民大学应用统计研究中心,北京100872 [2]中国人民大学统计学院,北京100872 [3]字节跳动数据推荐组,北京100024

出　　处：《系统工程理论与实践》2024年第5期1561-1576,共16页Systems Engineering-Theory & Practice

基　　金：教育部人文社会科学重点研究基地重大项目(22JJD910002);国家自然科学基金(72371241,72171229);全国统计科学研究项目(2022LD06)。

摘　　要：新闻推荐是一种重要的推荐场景,其推荐效果依赖于对新闻文本信息的充分挖掘.近年来,图神经网络因其强大的高阶信息挖掘能力,在推荐领域受到了广泛关注.然而,在新闻推荐领域,鲜有研究使用异质图神经网络,而且现有的异质图推荐模型也存在信息损失问题.为了充分挖掘新闻推荐场景中新闻、用户、主题、实体、类别等之间的高阶信息,更充分的挖掘新闻的文本特征,本文提出针对新闻推荐场景的多阶元路径引导的异质图神经网络推荐模型(简称MPNRec).该模型通过构建含有更多类型节点和边的异质图充分挖掘高阶信息,从而提高推荐效果.该方法在MIND small和Adressa 1week两个公开数据集上应用,较现有各种推荐方法在各项评价指标上至少能达到2%到5%的相对提升.News recommendation is an important recommendation scenario,and its effectiveness relies on the thorough exploration of news textual information.In recent years,graph neural networks(GNNs)have gained widespread attention in the field of recommendation due to their powerful ability to mine higher-order information.However,there is limited research on the use of heterogeneous graph neural networks in the field of news recommendation,and existing heterogeneous graph recommendation models also suffer from the problem of information loss.In order to fully exploit the high-level information among news,users,textual topics,entities,and categories in the news recommendation scenario,we propose a meta-path guided neighbors interaction recommendation model(MPNRec)for news recommendation.The MPNRec model builds a heterogeneous graph with more types of nodes and edges fully mine high-level information and improve the performance of news recommendation.On two public datasets(i.e.,MIND small and Adressa 1week),the MPNRec model can reach at least a 2%to 5%improvement in recommendation accuracy when compared with state-of-the-art methods.

关 键 词：新闻推荐 异质图神经网络 元路径 注意力机制 

分 类 号：TP18[自动化与计算机技术—控制理论与控制工程]