融合属性项情感的异质图神经网络产品推荐模型  

Heterogeneous Graph Neural Network with Attribute Sentiment for Product Recommendation

作　　者：李旸 符玉杰 王素格[3,4] 郑建兴 LI Yang;FU Yujie;WANG Suge;ZHENG Jianxing(School of Finance,Shanxi University of Finance and Economics,Taiyuan,Shanxi 030006,China;School of Computer Science and Technology,Taiyuan University of Science and Technology,Taiyuan,Shanxi 030024,China;School of Computer and Information Technology,Shanxi University,Taiyuan,Shanxi 030006,China;Key Laboratory Computational Intelligence and Chinese Information Processing of Ministry of Education,Shanxi University,Taiyuan,Shanxi 030006,China)

机构地区：[1]山西财经大学金融学院,山西太原030006 [2]太原科技大学计算机科学与技术学院,山西太原030024 [3]山西大学计算机与信息技术学院,山西太原030006 [4]山西大学计算智能与中文信息处理教育部重点实验室,山西太原030006

出　　处：《中文信息学报》2024年第12期137-147,共11页Journal of Chinese Information Processing

基　　金：国家自然科学基金(62106130,62076158,62376143,62072294);山西省基础研究计划项目(20210302124084),山西省高等学校科技创新项目(2021L284);山西省重点实验室开放课题基金(CICIP2023006)。

摘　　要：建立用户兴趣与产品属性项间的关联关系,有助于提升产品推荐系统中的准确度,因为在产品评论数据中,属性项可以作为构建用户与产品间关系的媒介。该文利用BERT-SAN/ChatGLM-Turbo自动获取评论数据的属性项情感信息,度量了用户对属性项的偏好度和属性项对产品口碑的贡献度,刻画了用户兴趣和产品口碑。在此基础上,以属性项为媒介构建了用户与产品关系的关联二部图,提出了融合属性项情感的多头注意力机制的异质图神经网络产品推荐模型,模型实现了图中边的更新和结点更新。最后,基于平均误差(MSE)训练了稳定的用户与产品属性项关联的二部图图神经网络产品推荐模型。在Yelp Restaurant和Digital Music两个数据集上进行了实验,验证了融合属性项情感的多头注意力机制的异质图神经网络产品推荐模型相比传统方法能够取得更好的性能,且属性项情感信息可以为产品推荐的可解释性提供依据。Establishing the relationship between user interests and product attributes term is helpful to improve the accuracy of product recommendation system.We propose a graph neural network product recommendation model based on a multi-head attention mechanism by integrating attribute term sentiment,which realizes the updating of edges and nodes in the graph.We utilize BERT-SAN/ChatGLM-Turbo to automatically obtain the attribute term sentiment information of the review data,and calculate the user s preference for the attribute term and the attribute s contribution to the product reputation.On this basis,a bipartite graph of the relationship between users and products is constructed with the attribute term.Finally,a stable user and product association bipartite graph neural network is trained in terms of MSE loss.Experiments on Yelp Restaurant and Digital Music datasets show that the proposed model is significantly better than state-of-the-art methods.

关 键 词：属性项情感分析 图神经网络 用户偏好度 属性项贡献度 可解释性 

分 类 号：TP391[自动化与计算机技术—计算机应用技术]