夏一雪 袁野 张文才 兰月新

〔摘 要〕[目的/意义]通过对网络舆情数据的动态监测和异常感知，及时预警舆情异常，为政府掌握舆情决策的先动优势提供理论模型和可行思路。[方法/过程]分析大数据环境下激增、波动等网络舆情数据异常现象，明确舆情趋势预测、动态感知异常等异常数据监测机理。基于此，首先运用Gompertz模型进行舆情趋势区间预测，其次定义偏离度进行数据异常评级，并确定预警等级，实现异常数据的及时捕捉和快速预警。[结论/结果]通过实例验证，证明了模型可行性，可以为政府舆情引导程度提供度量依据，也为编制智能化的舆情监测软件提供算法支持。

〔关键词〕大数据；网络舆情；异常数据；监测；预测

DOI：10.3969/j.issn.1008-0821.2018.06.012

〔中图分类号〕C912.6 〔文献标识码〕A 〔文章编号〕1008-0821（2018）06-0080-06

〔Abstract〕[Purpose/Significance]Through dynamic monitoring and abnormal perception of network public opinion data，it made early warning of abnormal network public opinion and provided theoretical models and practicable methods for grasping the preemptive advantage of network public opinion decision.[Method/Process]It analysed the abnormal phenomena of network public opinion data such as skyrocketing data and trend fluctuation under big data environment，made clear the monitoring mechanism of abnormal data，such as trend prediction，dynamic perception of abnormal data，etc.On this basis，the first step was using Gompertz model to predict the trend of network public opinion theoretical interval.Secondly，the deviation degree was defined for data anomaly rating，and the early warning level was determined to realize the timely capture and rapid early warning of abnormal data.[Result/Conclusion]The feasibility of the model was proved by an example，which could provide the measurement basis for the guidance degree of network public opinion，and also provided the algorithm support for compiling intelligent monitoring software of network public opinion.

〔Key words〕big data；network public opinion；abnormal data；monitoring；prediction

1 現状分析

根据第41次《中国互联网络发展状况统计报告》显示，截至2017年12月，我国手机网民规模达7.53亿，网民中使用手机上网人群的占比由2016年的96.1%提升至97.5%[1]。随着移动宽带互联网的普及，热点舆情以及由其引发的舆情反转、衍生舆情等各类网络舆情事件层出不穷，上海外滩踩踏事故（2014）、南海仲裁（2016）、魏则西事件（2016）等舆情信息数量激增，哈尔滨天价鱼（2016）、李文星事件（2017）、杭州保姆纵火（2017）等舆情的急速反转和剧烈波动，都蕴含着大量的网络舆情数据异常变化情况，加之受网络水军、网络推手、舆论战等影响，导致网络舆情异常现象频出。在舆情监测过程中，相较于常态舆情监测，异常数据监测的决策支持价值更加突出，特别是在数据异常变化初期，及时监测并提前预警，有助于政府掌握舆情决策的先动优势。……