以用户为中心的微博信息转发研究与预测

发布时间：2018-07-29 09:31

【摘要】：微博信息转发作为信息传播研究的关键问题之一。预测信息的转发概率及传播趋势在信息传播、舆情监控、产品推荐等方面具有重要的应用价值。现有研究主要基于网络结构及信息的历史传播规律预测信息的未来传播趋势,大多忽略了用户间的个体差异。在基于用户行为的转发预测中还主要是站在信息发布者的角度研究信息的被转发因素,较少研究用户转发信息的影响因素。本文主要以用户为中心,站在信息接收者的角度,通过挖掘影响用户转发的主要因素并结合机器学习中分类算法进行预测,主要工作如下:首先,根据实际问题需求通过微博平台提供的API抓取研究所需要的数据集,包括用户信息、微博信息、用户关系信息和转发关系信息等,并对数据集的特征及完整性进行分析和描述,并结合实际特征的影响情况进行选取。。然后,挖掘影响用户转发行为的重要因素,包括信息发布者特征、信息接收者特征及用户间交互特征,通过挖掘特征与转发之间的关系图分析所选特征的特点及影响。最后,使用支持向量回归、朴素贝叶斯及随机森林三个分类算法并结合信息转发的影响因素,对用户是否转发信息进行预测,通过实验对比结果选取最适合模拟网络中真实转发过程的分类算法。通过模型分析证实了挖掘用户特征对信息转发行为预测研究的必要性,运用误分率得出不同因素影响信息转发行为的重要程度。
[Abstract]:Weibo message forwarding is one of the key issues in the research of information dissemination. Forecasting the probability and trend of information forwarding has important application value in information dissemination, public opinion monitoring, product recommendation and so on. The existing research mainly based on the network structure and the information history dissemination law predicts the information future dissemination tendency, mostly ignores the individual difference between the user. In user behavior based forwarding prediction, we mainly study the factors of information being forwarded from the point of view of information publisher, and less study the influencing factors of user forwarding information. In this paper, user-centered, from the perspective of information receiver, by mining the main factors that affect the user forwarding and combining with the classification algorithm in machine learning, the main work is as follows: first, According to the demand of practical problems, the data set of the research is captured by API provided by Weibo platform, including user information, Weibo information, user relationship information and forwarding relation information, and the characteristics and integrality of the data set are analyzed and described. And combined with the actual characteristics of the impact of the selection. Then, the important factors that affect the user's forwarding behavior are mined, including the information publisher feature, the information receiver feature and the user interaction feature, and the characteristics and effects of the selected features are analyzed by mining the relationship graph between the features and the forwarding. Finally, support vector regression, naive Bayes and random forest classification algorithms are used to predict whether the information is forwarded or not. By comparing the experimental results, the classification algorithm which is most suitable for simulating the real forwarding process in the network is selected. The necessity of researching the prediction of information forwarding behavior by mining user features is confirmed by model analysis, and the importance of different factors affecting information forwarding behavior is obtained by using the error rate.
【学位授予单位】：首都经济贸易大学
【学位级别】：硕士
【学位授予年份】：2017
【分类号】：G206

【参考文献】