社交媒体内容流行度的建模与预测方法研究
文献类型：学位论文
| 作者 | 孔庆超
|
| 学位类别 | 工学博士 |
| 答辩日期 | 2016-05-29 |
| 授予单位 | 中国科学院研究生院 |
| 授予地点 | 北京 |
| 导师 | 毛文吉 |
| 关键词 | 用户生成的内容 内容流行度 流行度预测 社交媒体分析 |
| 中文摘要 | 近年来，随着互联网的迅猛发展和普及，社交媒体已经成为人们获取信息、表达思想以及分享观点和情感的重要平台。社交媒体中用户间的交互行为使得某些用户生成的内容（如讨论帖、微博话题等）变得流行。对所关注内容的流行度进行建模和预测在商业、安全等多个领域中具有十分重要的研究和应用价值。本论文工作聚焦社交媒体内容的流行度建模和预测问题，研究建立刻画其动态演化过程的模型来预测其流行度的发展趋势和发展阶段，并研究不同媒体内容之间的相互作用对流行度建模和预测的影响。本论文采用多个典型的社交媒体形式，包括豆瓣小组和新浪微博的数据对所提出的流行度建模和预测方法进行了有效性验证。 本论文的主要贡献包括： 1) 面向网络社交媒体内容，给出了其流行度趋势预测问题定义；总结了影响流行度的相关动态因素，包括评论树结构特性和用户回复关系；利用局部特性计算各个动态因素的先验信息并改进kNN算法，提出一种融合多个动态因素的流行度动态演化趋势预测方法，并通过实验验证了该方法的有效性； 2) 在流行度趋势预测的基础上，给出了流行度不同发展阶段的定义并提出流行度发展阶段预测问题；面向多分类问题给出一种基于动态因素的流行度阶段预测方法；同时针对典型社交媒体形式，挖掘其流行度演化模式并提出一种结合演化模式的流行度阶段预测方法，并通过实验验证了该方法的有效性； 3) 首次研究并提出基于社交媒体内容之间相互作用关系的流行度预测方法；基于社交媒体内容的交互特点，给出计算媒体内容相互作用关系的算法；利用非负矩阵分解从相互作用信息中自动抽取相互作用特征，建立基于社交媒体内容相互作用特征的流行度预测方法，并通过实验验证了该方法的有效性。 |
| 英文摘要 | With the rapid development and prevalence of the Internet in recent years, social media has become important platforms for Web users to acquire information, express thoughts, share opinions and feelings. The interaction behaviors between social media users make some online contents popular, for example, threads and topic hashtags. The popularity modeling and prediction of online contents are of great research and application importance in domains such as business and security. Focusing on the popularity modeling and prediction problem of online contents in social media, we build models that describe the dynamic evolution of popularity and predict its trend and evolution stages, and explore the influence of the interactions between different online contents on popularity prediction performance. We also evaluate the effectiveness of the proposed popularity prediction methods by using data from multiple social media platforms, including Douban group and Sina Weibo. The main contributions of this thesis are as follows: 1) We first address the popularity trend prediction problem with respect to the online contents of social media. We summarize some related dynamic factors that influence the popularity trend, including the structural properties of the comment tree and user reply network. We first calculate the prior information of the dynamic factors using the locality property, and propose an algorithm that combines the prediction performance of multiple dynamic factors based on kNN. Experimental studies verify the effectiveness of the proposed method. 2) On the basis of popularity trend prediction, we first give the definitions of different popularity stages and propose the problem definition of popularity stage prediction. We then propose a popularity stage prediction method based on dynamic factors with respect to a multiclass classification problem. The proposed method first extracts the popularity evolution patterns of some typical social media online contents and incorporates evolution patterns for popularity stage prediction. Experimental studies verify the effectiveness of the proposed method. 3) We first propose a popularity prediction method based the interactions of different online contents. Specifically, we first propose an algorithm which calculates the interaction relation of online contents based on the interactive characteristics of online contents in social media. Then we adopt the non-negative matrix factorization technique to extract interaction features from the interaction matrix and build popularity prediction models using the interaction features. Experimental studies verify the effectiveness of the proposed method. |
| 语种 | 中文 |
| 源URL | [http://ir.ia.ac.cn/handle/173211/11581]  |
| 专题 | 毕业生_博士学位论文 |
| 作者单位 | 中国科学院自动化研究所 |
| 推荐引用方式 GB/T 7714 | 孔庆超. 社交媒体内容流行度的建模与预测方法研究[D]. 北京. 中国科学院研究生院. 2016. |
入库方式： OAI收割
来源：自动化研究所
浏览0
下载0
收藏0
其他版本
除非特别说明，本系统中所有内容都受版权保护，并保留所有权利。