公共卫生事件中的微博舆情趋势预测研究

发布时间：2018-06-06 05:18

本文选题：公共卫生事件 + 微博舆情　；参考：《哈尔滨工业大学》2017年博士论文

【摘要】：微博等社交媒体给危机沟通增添了复杂性。通过微博,信息能够被快速地发布、转发,使得受众数目成指数级增长。这一特性使公共卫生事件微博舆情快速形成,导致管理者对一些公共问题产生错误或滞后的反应,后果是增加公众的不安全感,使得原本在传统传播模式下可以控制的公共卫生事件网络舆情进一步升级,甚至形成社会危机。显然,能及时为公共危机管理部门提供舆情趋势信息,帮助他们提早发现会引起公众广泛关注和热烈讨论的舆情热点微博是非常重要的。本研究进行了公共卫生事件中的微博舆情信息采集、分析和舆情趋势预测方法研究。研究成果可以帮助危机管理部门及时判断公共卫生事件微博舆情趋势,进行高效、精准的危机沟通与处理,有助于促进社会的和谐发展。论文主要的研究内容如下:首先,提出了公共卫生事件中的微博舆情信息采集方法。舆情信息采集的第一个步骤是利用新闻媒体数据源构筑面向公共危机沟通的微博舆情信息采集过滤流程。然后,基于Word2Vec技术将语料中的词汇转化为高维向量空间中的向量,通过计算词向量之间的距离,提取公共卫生事件关键词,并利用得到的关键词采集公共卫生事件相关微博。最后,利用“80-20”法则界定舆情热点微博,从而发现需要公共管理者高度关注和有针对性地进行危机沟通的微博舆情信息。其次,分析公共卫生事件微博舆情趋势的影响因素。研究基于“启发-分析双处理理论”、“劝说理论”、“社会影响理论”,从微博内容特征、微博作者特征、以及微博社会影响特征三个方面,对微博舆情趋势影响因素进行分析、提取和测量。在此基础上,计算每一个微博特征对舆情趋势的信息增益,得出不同微博特征对舆情趋势的影响。保留对舆情趋势影响比较大的微博特征,使得舆情趋势预测模型能够集中到最富含预测信息的少量特征上,提高预测模型的有效性。然后,对影响舆情趋势的微博特征进行粗糙集约简。针对影响舆情趋势的微博特征中可能还存在一些冗余和具有不确定性的特征,利用粗糙集特征约简方法,在不降低建模精度和预测能力的前提下,得到在近似质量最优意义下的最小约简特征集合,简化建模过程和所建模型的复杂性,有利于从特征选择上进一步提高舆情趋势预测的准确度。同时,粗糙集方法还能提取公共卫生事件微博舆情趋势的规则作为预报因子,为舆情爆发预警奠定基础。最后,建立公共卫生事件舆情趋势预测的集成分类模型。研究基于粗糙集特征约简后得到的最小约简特征集合,构建了微博舆情趋势的集成分类预测模型。对863条有关公共卫生事件的微博信息进行的实证分析证实了本研究提出的集成分类模型可以有效预测公共卫生事件微博舆情趋势。
[Abstract]:Social media such as Weibo add complexity to crisis communication. Through Weibo, information can be quickly distributed and forwarded, increasing the number of audiences exponentially. This characteristic causes public health events Weibo public opinion to form quickly, causes the manager to produce the mistake or the lag response to some public problem, the consequence is to increase the public insecurity, Public health events can be controlled in the traditional mode of public health events network public opinion to further upgrade, and even form a social crisis. Obviously, it is very important to provide the public crisis management department with the public opinion trend information in time and help them to find the public opinion hot spot Weibo which will arouse the public attention and the heated discussion in advance. In this study, the Weibo public opinion information collection, analysis and public opinion trend prediction method were studied. The research results can help the crisis management department judge the public health event Weibo public opinion trend in time, carry on the efficient, accurate crisis communication and deal with, help to promote the harmonious development of the society. The main contents of this paper are as follows: firstly, the method of collecting Weibo public opinion information in public health events is proposed. The first step of collecting public opinion information is to construct the Weibo public opinion information collection and filtering flow based on the news media data source for public crisis communication. Then, based on the Word2Vec technology, the vocabulary in the corpus is transformed into the vector in the high-dimensional vector space. By calculating the distance between the word vectors, the public health event keywords are extracted, and the public health event related Weibo is collected by the obtained keywords. Finally, the "80-20" rule is used to define the hot Weibo of public opinion, and the Weibo public opinion information which needs the attention and pertinence of public administrators to communicate crisis is found. Secondly, the influencing factors of Weibo public opinion trend of public health events are analyzed. The research is based on "heuristic and analytical double processing theory", "persuasion theory", "social impact theory", from three aspects: Weibo content characteristics, Weibo author characteristics, and Weibo social impact characteristics. The influencing factors of Weibo public opinion trend are analyzed, extracted and measured. On this basis, the information gain of each Weibo feature to the public opinion trend is calculated, and the influence of different Weibo features on the public opinion trend is obtained. The Weibo features which have a great influence on the trend of public opinion are retained so that the prediction model of public opinion can concentrate on a small number of features which are most rich in prediction information and improve the effectiveness of the prediction model. Then, the Weibo features which affect the trend of public opinion are reduced by rough set. In view of some redundant and uncertain features in the Weibo features that affect the trend of public opinion, the rough set feature reduction method is used to reduce the modeling accuracy and prediction ability. The minimum reduction feature set in the sense of the best approximate quality is obtained to simplify the modeling process and the complexity of the established model, which is conducive to further improving the accuracy of prediction of public opinion trend from feature selection. At the same time, the rough set method can extract the rules of public health event Weibo public opinion trend as a forecast factor, and lay a foundation for public opinion outbreak early warning. Finally, an integrated classification model of public health event trend prediction is established. Based on the minimum reduction feature set obtained from feature reduction in rough set, an integrated classification and prediction model of Weibo public opinion trend is constructed. The empirical analysis of 863 Weibo information on public health events proves that the integrated classification model proposed in this study can effectively predict the trend of public health events Weibo public opinion.
【学位授予单位】：哈尔滨工业大学
【学位级别】：博士
【学位授予年份】：2017
【分类号】：G206

【相似文献】