于海¹，吕晴晴²，时鹏³，王铮¹，胡长军¹
AuthorsHTML:于海¹，吕晴晴²，时鹏³，王铮¹，胡长军¹
AuthorsListE:Yu Hai¹，Lü Qingqing²，Shi Peng³，Wang Zheng¹，Hu Changjun¹
AuthorsHTMLE:Yu Hai¹，Lü Qingqing²，Shi Peng³，Wang Zheng¹，Hu Changjun¹
Unit:1. 北京科技大学计算机与通信工程学院，北京 100083；
2. 中国电子科技集团公司第十五研究所，北京 100083；
3. 北京科技大学国家材料服役安全科学中心，北京 100083
Unit_EngLish:1. School of Computer & Communication Engineering，University of Science and Technology Beijing，Beijing 100083，China；
2. China Electronics Technology Group Corporation 15th Research Institute，Beijing 100083，China；
3. National Center for Materials Service Safety，University of Science and Technology Beijing，Beijing 100083，China
Abstract_Chinese:随着近些年新一代信息网络技术的高速发展，人们可以通过社交网络平台更快、更广地获知社会上的各类事件，这使得事件在社交网络中的传播逐步呈现出高速化、扩大化的特点．针对这种情况，为了更好地管理社交网络中的事件，提高对网络事件信息的治理水平，有必要对社交网络信息传播进行分析．流行度预测是在线社交网络事件信息传播分析中的研究重点．对事件流行度的预测能够为网络事件发生、发展、高峰、终结等提供深刻的见解．尽管流行度预测已经被广泛研究，但是事件相关信息与流行度相关联的因素缺少即时可用的指标数据、指标数据差异化等问题使得有效地预测事件流行度仍然没有得到较好的解决．有鉴于此，本文设计了一种基于在线社交网络事件库多因素耦合的流行度预测方法．具体来说，首先提出了一种基于社交网络事件库的多因素指标获取方法，利用事件库对于社交网络数据的统一存储，从多源异构数据中提取各因素指标．其次提出了一种多因素耦合的流行度预测方法，通过分组嵌入得到因素指标的可相互结合的低维表示，在预测中实现多因素指标的综合利用．最后将Twitter7数据集中3000个主题标签包含的推文作为实验对象进行平均准确率计算．实验结果表明：与已有的深层神经网络模型(DNN)、支持向量回归模型(SVR)、SH流行度预测模型等相比，本研究所提出预测方法在预测准确度上具有明显的优越性．
Abstract_English:With the rapid development of the new generation of information network technology in recent years，people become aware of all kinds of social events rapidly and extensively through the social network platform，which speeds up and expands the diffusion of events in the social network. Because of this situation，to manage the events in the social network and improve the governance level of network event information effectively，information diffusion in the social network needs to be analyzed. Popularity prediction is the focus of online social network event information diffusion analysis. Popularity prediction can provide profound insights into the occurrence，development，peak，and end of network events. Although popularity prediction has been widely investigated，the lack of instant and available indicator data for factors associated with event-related information and popularity and the difficulty in differentiating indicator data hinder the effective prediction of event popularity. For this reason，this study designs a popularity prediction method based on the multi-factor coupling of the online social network event base. First，a multi-factor indicator acquisition method based on the social network event database，which uses the event database to uniformly store social network data and extract each factor indicator from multi-source heterogeneous data，is proposed. Second，a multi-factor coupling method for popularity prediction is proposed，with the low-dimensional representation of factor indicators that can be combined is obtained by grouping and embedding to realize the comprehensive utilization of multi-factor indicators during prediction. Finally，the tweets contained in 3 000 subject tags in the Twitter 7 data set are utilized as the experimental subjects to calculate the average accuracy. The experimental results show that，compared with the existing deep neural network，support vector regression，and SH popularity prediction models，the prediction method proposed in this study has superior prediction accuracy.
Keyword_Chinese:流行度预测；多因素耦合；累积性因素；固有性因素
Keywords_English:popularity prediction；multi-factor coupling；cumulative factor；inherent factor

PDF全文下载地址:http://xbzrb.tju.edu.cn/#/digest?ArticleID=6558