面向网络舆情的无监督演化趋势评估方法

Research on Unsupervised Method for Changing Trend Evaluation of Internet Public Opinion Events

针对网络舆情突发性强、标注数据较少且管控资源有限等问题,提出了一种网络舆情演化趋势评估无监督学习算法,筛选演化趋势重要的舆情事件进行优先管控,提升网络监管的工作效率。针对舆情事件并发性强的特点,利用多指标排序算法对舆情演化趋势重要性评估问题进行形式化描述;针对舆情突发性强、难以获得大量标注数据的问题,利用主曲线排序算法对舆情演化趋势重要性评估问题进行建模,采用3阶贝塞尔曲线进行模型求解,充分利用评估指标中的顺序关系和数值关系;结合典型公开数据集和自主构建的舆情数据集对所提算法进行了验证分析,实验结果表明,所提算法可在无标注数据的情况下实现舆情事件演化态势重要性的评估,为资源有限情况下的舆情事件管控提供决策支撑。

Aiming at the abruptness posed by public opinion emergencies,limited labeled data and management resources,we proposed an unsupervised algorithm for evaluating the importance of public opinion changing trend and controlling the important public opinion events within limited time and reducing their harmful influence to the society.Firstly,the importance of changing trend evaluation problem is transformed into a multi-index ranking based on the management experience.Secondly,to solve the problems posed by limited labeled data,we use the principal curve algorithm to formulate the problem of changing trend evaluation,and use the third-order Bessel curve to obtain the final ranking results.The designed model can fully capture the structural and value characteristics of the original data.Finally,we employ the typical public data set and the self-built public opinion event data set to verify the proposed method.The experimental results verify that the developed method has high efficiency in public opinion event changing trend ranking without prior knowledge,and provides a decision support for public opinion event management with limited resources.