社交网站中用户评论行为预测

发布时间：2018-02-28 04:33

本文关键词： 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　出处：《智能系统学报》2015年03期 　论文类型：期刊论文

【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 very convenient platform for users to communicate with each other and express their opinions and opinions. Modeling and forecasting user behavior on social networking sites is of great social significance and practical value in many fields, such as security, business, etc. In recent years, researchers have paid more and more attention to the behavior of users in social networking sites, predicting whether users will participate in the discussion, and using feature-based machine learning methods, which include discussion posts and their content. User behavior characteristics and social relationships, and the introduction of parameter control data set imbalance. The experiment uses real data from the Douban group. The experimental results show that, The new features of user behavior and social relationships, as well as the method of dealing with unbalanced data sets, can effectively improve the prediction effect of user comment behavior. It is further explained that the users' historical behavior and their social network have great influence on the current comment behavior.
【作者单位】：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;
【基金】：国家自然科学基金资助项目(61175040,U1435221)
【分类号】：TP181;TP393.092

【共引文献】