社交网站中用户评论行为预测

PDF下载HTML

Title:

User comment behavior prediction in social networking sites

作者:

孔庆超, 毛文吉, 张育浩

中国科学院自动化研究所 复杂系统管理与控制国家重点实验室, 北京 100190

Author(s):

KONG Qingchao, MAO Wenji, ZHANG Yuhao

State Key Laboratory of Management and Control for Complex Systems, Institute of Automation, Chinese Academy of Science, Beijing 100190, China

关键词:

社交网络; 用户评论; 机器学习; 行为建模; 行为预测; 不平衡性数据集

Keywords:

social network; user comments; machine learning; behavior modeling; behavior prediction; imbalance dataset

分类号:

TP391

DOI:

10.3969/j.issn.1673-4785.201403019

文献标志码:

A

摘要:

社交网站为用户相互交流、发表意见和观点提供了非常便利的平台.对社交网站的用户行为进行建模和预测对于安全、商业等多个领域具有十分重要的社会意义和应用价值,近年来逐渐得到研究者的重视.面向社交网站中用户评论行为,预测用户是否会参与讨论.采用基于特征的机器学习方法,其中特征包括讨论帖子及其内容、用户行为特征和社交关系,并引入参数控制数据集的不平衡性.实验采用来自豆瓣小组的真实数据.实验结果表明,新提出的用户行为和社交关系特征以及对不平衡数据集的处理方法能够有效提高用户评论行为的预测效果,进一步说明用户的历史行为和所在的社交关系网络对当前的评论行为有较大影响.

Abstract:

Social networking sites provide a convenient way for users to communicate with others and to present opinions. Related researches on modeling and predicting user behaviors in social networking sites are of vital importance for many applications in the domains of security and business. The aim of this paper is to predict user comment behavior based on postings in social networking sites. A feature-based machine learning approach is employed, which includes features from the postings, content, user behaviors and social relations, and introduces a parameter to control the imbalanceness of the dataset. Real-world datasets from Douban Group were used in the experiments. The experimental results showed that the user behavior and social relation features and the imbalance processing technique effectively improved the prediction performance of user comment behaviors. This further demonstrates that the user comment behavior is largely affected by their behavior history and social network.