基于图神经网络和异构信息的兴趣点分类  

Graph Convolution Networks for POI Classification with Heterogeneous Information

作　　者：任成森 杨易扬 郝志峰[1,2] REN Cheng-sen;YANG Yi-yang;HAO Zhi-feng(College of Computers,Guangdong Universily of Teclmology,Guangzhou 510006;College of Matliematics&Big Data.Foshan University,Guangzhou 528225)

机构地区：[1]广东工业大学计算机学院,广州510006 [2]佛山科学技术学院,佛山528225

出　　处：《现代计算机》2021年第6期3-7,15,共6页Modern Computer

基　　金：国家自然科学基金(No.61603101)、国家自然科学基金(No.61472089);NSFC-广东联合基金(No.U1501254);广东省自然科学基金资助项目(No.2014A030308008);广东省科技计划项目(No.2015B010108006)

摘　　要：基于移动网络的快速发展,饿了么、美团等这一类基于位置的社交网络在我们日常生活中日益普及,随即积累了大量的用户在兴趣点的签到数据、评论这就激发了许多研究者对兴趣点进行各种研究,例如兴趣点分类、推荐任务但现有众多的研究都是基于评论数据,进而把兴趣点分类转化为文本分类,而忽略兴趣点的其他信息,例如兴趣点的名称信息。针对以上问题,提出结合兴趣点名称和评论文本的信息,为城市兴趣点构建一个异构图网络,利用图卷积神经网络模型进行兴趣点分类:实验证明,所提出的异构图模型,充分利用这两种异质信息,并相对于基准模型不仅保持其准确率,还提高训练性能,在现实世界的真实数据中有着良好的表现。Based on the rapid development of mobile networks, location-based social networks such as Ele.me and Meituan have become increasingly popular in our daily lives, and a large number of users’ check-in data and comments at points of interest have been accumulated. This motivates many researchers to conduct various researches on points of interest, such as point of interest classification and recommendation tasks. However, many existing researches are based on comment data, and then convert the classification of points of interest into text classification, while ignoring other information of the points of interest, such as the name information of the points of interest. In response to the above problems, this paper proposes to combine the information of the names of points of interest and the review text to construct a heterogeneous graph network for the city points of interest, and use the graph convolutional neural network model to classify the points of interest.Experiments show that the heterogeneous graph model proposed in this paper makes full use of these two types of heterogeneous information, and not only maintains its accuracy compared with the benchmark model, but also improves the training performance, and has a good performance in real data in the real world.

关 键 词：图卷积神经网络 兴趣点分类 异构网络 鲁棒性 

分 类 号：TP391.1[自动化与计算机技术—计算机应用技术] TP183[自动化与计算机技术—计算机科学与技术]