基于中文微博的热点事件情感倾向分析

发布时间：2018-06-22 15:13

本文选题：情感分类 + 社交网络　；参考：《北京邮电大学》2015年硕士论文

【摘要】：近年来,微博成为了人们非常重要的在线网络活动平台。很多热点事件发生后用户能够在微博上获取相关消息,并且通过微博平台提供的发布、转发、评论等功能十分方便的参与到热点事件的讨论与传播过程中。微博平台开放的机制使得用户可以在微博上对不同主题热点事件发表观点、表达情感倾向。对热点事件相关微博数据进行情感分析,可以从中获取大众对具体事件的整体情感状态和情感传播转换情况,这些信息对舆情分析、营销效果评估等实际应用有重要的指导意义。本文对中文微博文本的短文本性和新颖性进行了分析,针对这类特性抽取情感学习的特征,构建了适合中文微博情感倾向分类的模型,并针对微博情感倾向的正负极性和中性的不同外延,提出并对比了一步分类和二步分类策略。在建立的情感分类模型的基础上,利用微博热点事件数据集对社会、娱乐、体育三类主题的典型热点事件在微博传播的激烈程度、参与度等方面的情感特性进行了分析。在此基础上,本文进一步通过典型热点事件传播数据集,分析了情感传播的动态过程和传播过程中的情感转变模式,描述了情感转换统一概率框架,并且基于条件随机场(CRF)构建情感转换预测模型,能够对事件微博传播中的情感传播转换进行预测。实验表明,本文构建的基于中文微博的情感分类模型对微博进行正向-中性一负向三种极性的分类,使用一步三分类策略的F1-值为74.9%,而二步分类策略的预测效果达到了82.4%,情感分类的效果较为理想,且验证了二步分类策略的有效性。情感动态传播中的基于条件随机场情感转换预测模型对情感转换的预测效果F1-值为60.2%,相较基准支持向量机(SVM)模型的效果提高3.7%,能够较好刻画情感转换传播网络,对于事件传播中情感的转换有一定的预测能力。
[Abstract]:In recent years, Weibo has become a very important online network platform. After a lot of hot events happen, users can get relevant information on Weibo, and participate in the discussion and propagation of hot events conveniently through the functions of publishing, forwarding, commenting and so on provided by Weibo platform. The open mechanism of Weibo platform enables users to express their opinions and emotional tendencies on Weibo on different topics and hot events. Through emotional analysis of Weibo data related to hot events, we can get the overall emotional state of the public on specific events and the situation of emotional communication conversion, which is the analysis of public opinion. Marketing effect evaluation and other practical applications have important guiding significance. This paper analyzes the short text nature and novelty of Chinese Weibo texts, extracts the characteristics of emotional learning from these features, and constructs a model suitable for the classification of Chinese Weibo affective tendencies. Aiming at the positive and negative pole and neutral extension of Weibo's affective tendency, the one-step classification and two-step classification strategy are proposed and compared. On the basis of the emotion classification model established, the emotional characteristics of the typical hot events of social, entertainment and sports topics in Weibo propagation and participation were analyzed by using the Weibo hot event data set. On this basis, this paper further analyzes the dynamic process and the mode of emotional transformation in the process of emotional communication through the typical hot event propagation data set, and describes the unified probabilistic framework of emotional transformation. Based on conditional Random Field (CRF), a prediction model of affective transformation is constructed, which can predict the emotional transition in event Weibo propagation. The experiment shows that the emotion classification model based on Chinese Weibo is used to classify Weibo with three polarities: forward, neutral and negative. The F1- value of one-step and three-step classification strategy is 74.9, while the prediction effect of two-step classification strategy is 82.4. The effect of emotion classification is satisfactory, and the validity of two-step classification strategy is verified. The prediction effect of conditional random field emotion transformation prediction model in dynamic emotion propagation is 60.20.The result is 3.7g higher than that of SVM model, which can depict affective transformation communication network. There is a certain ability to predict the change of emotion in event communication.
【学位授予单位】：北京邮电大学
【学位级别】：硕士
【学位授予年份】：2015
【分类号】：TP391.1

【参考文献】