面向社交媒体的用户建模方法研究
蔡驰宇
Subtype博士
Thesis Advisor曾大军
2019-05-25
Degree Grantor中国科学院大学
Place of Conferral中国科学院自动化研究所
Degree Name工学博士
Degree Discipline社会计算
Keyword社交媒体 用户建模 用户检测 跨平台 深度学习
Abstract

随着互联网特别是移动互联网的高速发展,社交媒体作为Web2.0应用的代表之一,得到了广大网民的认可和青睐。社交媒体的蓬勃发展和广泛应用,使得其对人类社会的影响日益变大,成为了当前网民获取信息、发布信息、参与线上活动的重要通道。在此背景下,通过挖掘网络用户在社交媒体平台中的相关数据,可以有效的分析用户的诸多特性,例如年龄、职业、兴趣、偏好、行为模式等。面向社交媒体的用户建模在政务、商业、教育、公共安全等领域有着十分重要的研究意义和实际应用价值。本论文聚焦面向社交媒体的用户建模问题,从在线行为、写作风格、跨平台等多个角度探索了用户建模方法,并研究其在用户检测方面的应用。本论文采用国内外主流社交媒体平台新浪微博、Twitter等作为实验数据源,对提出的模型和方法进行了有效性验证。
本论文的主要研究内容和贡献包括:
1.探索了深度神经网络在用户建模领域的应用,首次提出了一种基于行为信息增强的深度模型。针对现有工作中缺乏对用户行为影响因素的分析,本论文结合组织行为学中个体行为影响因素理论,分析内生因素(人体生物节律)和外生因素(社会文化因素等)对用户在线行为的影响,提出了一种基于内外生因素的在线行为建模方法。在此基础上,融合时序文本内容信息和在线行为信息,提出了一种基于在线行为的用户建模方法,并用于社交机器人检测。实验结果表明所提出方法在该任务上优于基准方法,且个体行为影响因素分析有助于提升用户在线行为建模效果。
2.首次聚焦于网络流行文化影响下的用户个性化写作模式研究。本论文研究了以用户生成内容为代表的网络流行文化现状,提出了一种基于潜在语义信息的网络情感派生词识别框架,以辅助用户生成内容的收集。在此基础上,提出了一种基于用户生成内容的写作模式建模方法。该方法设计了一种注意力机制用于关注用户生成内容对于写作模式的影响,并提出了一种记忆网络用于挖掘用户个性化写作模式。进一步本论文融合多类特征信息,提出了一种基于个性化写作模式的用户建模方法,并用于社交媒体用户识别和社交机器人检测中。实验结果验证了所提出的写作模式建模方法效果优于多个现有方法,且基于个性化写作模式的用户建模方法在相应任务中表现更优。
3.针对现有跨平台用户建模工作缺乏用户动态信息的分析和挖掘,提出了一种基于动态信息的跨平台用户建模方法。该方法从情感波动、在线行为和写作模式三个角度对用户动态特征进行分析和建模。其后,融合传统特征信息和动态特征信息,构建深度匹配网络,实现跨平台用户建模与检测。在真实的多社交平台数据集上,实验验证了所提方法相比于多个现有方法有较大的提升,证明了所提出方法的有效性。

Other Abstract

With the rapid development of the Internet and mobile Internet, social media, as one of the representatives of Web2.0 applications, has been recognized and favored by the majority of netizens. The vigorous development and wide application of social media has made its influence on human society increasingly larger. Social media has become an important channel for netizens to obtain information, post information and participate in online activities. By mining data from social media users, many characteristics of users can be effectively analyzed, such as age, occupation, interest, preferences, behavior patterns and so on. User modeling for social media has important research significance and practical application value in the fields of government affairs, business, education and public security. In this thesis, we focus on user modeling for social media, explore user modeling methods from multiple perspectives including online behavior, writing style, cross platform, and research its application in user detection. We carry out experimental studies to evaluate the effectiveness of the proposed user modeling and detection methods on social media datasets, including Sina Weibo dataset, Twitter dataset, etc.
The major works and contributions of this thesis are summarized as follows:
1. We explore the application of deep neural network in the field of user modeling, and first propose a behavior information enhance deep model. In view of the lack of research on the factors affecting user behavior in the existing works, combining the theory of influencing factors of individual behavior in Organizational Behavior Science, this thesis analyzes the effects of endogenous factors (human biological rhythms) and exogenous factors (social and cultural factors) on users' online behavior, and propose an endogenous and exogenous factors based online behavior model. On this basis, we propose an online behavior based social user model which fuse temporal text information and online behavior information, and implement social bot detection. We finally conduct experimental studies to verify the effectiveness of proposed methods.
2. We first focus on the user personalized writing pattern under the influence of network popular culture. This thesis first studies the related works of user-generated content, and propose a latent semantic based framework for web-derived emotional word detection which help to collect user-generated content. Based on this, we propose a user-generated content attention based writing pattern model, which focus on the influence of user-generated content on writing mode through the attention mechanism, and constructs the memory network to mine the users’ personalized writing patterns. In addition, we propose a personalized writing pattern based social user model which fuse user basic information, content information and writing pattern information, and implement social user identification and social bot detection. The experimental results verify that the proposed writing pattern model is more effective than many existing methods, and the proposed social user model performs better in corresponding tasks.
3. In view of the lack of research on the user dynamic information for existing cross-platform user modeling works, we propose a dynamic information based cross platform user modeling method. We first analyze and model three dynamic features including emotional fluctuation, online behavior and writing pattern. Then, we integrate traditional information and dynamic information. Finally, we build a deep matching network and realize cross-platform user modeling and detection. In real-world multiple social platforms datasets, experiments verify that the proposed method has a great improvement compared with many existing methods, and proves the effectiveness of the proposed method.

Pages112
Language中文
Document Type学位论文
Identifierhttp://ir.ia.ac.cn/handle/173211/23796
Collection复杂系统管理与控制国家重点实验室_互联网大数据与信息安全
Recommended Citation
GB/T 7714
蔡驰宇. 面向社交媒体的用户建模方法研究[D]. 中国科学院自动化研究所. 中国科学院大学,2019.
Files in This Item:
File Name/Size DocType Version Access License
面向社交媒体的用户建模方法研究.pdf(4240KB)学位论文 开放获取CC BY-NC-SAApplication Full Text
Related Services
Recommend this item
Bookmark
Usage statistics
Export to Endnote
Google Scholar
Similar articles in Google Scholar
[蔡驰宇]'s Articles
Baidu academic
Similar articles in Baidu academic
[蔡驰宇]'s Articles
Bing Scholar
Similar articles in Bing Scholar
[蔡驰宇]'s Articles
Terms of Use
No data!
Social Bookmark/Share
All comments (0)
No comment.
 

Items in the repository are protected by copyright, with all rights reserved, unless otherwise indicated.