利用Hurst指数预测网络话题热度  

Hot Trend Prediction of Network Topic Based on the Hurst Exponent

作　　者：程葳[1] 

机构地区：[1]北京城市学院研究部,北京100083

出　　处：《现代计算机（中旬刊）》2012年第8期3-7,共5页Modern Computer

基　　金：国家高技术研究发展计划(863)项目(No.2005AA147030);北京市教育委员会科技发展计划面上项目(No.KM200600006002)

摘　　要：面向大规模真实数据,应用聚合方差法、周期图和重标极差法对互联网信息发布数据进行研究,发现网民发布信息具有自相似和长相关特性。通过对各序列的Hurst指数估计,揭示了网络论坛日发帖量未来变化与历史的变化趋势一致;网民发帖量低发帖量网民与高发帖量网民的变化趋势一致;以及每日网民发帖量后发帖网民与先发帖网民的变化趋势一致等特性。这些性质对深入认识网络论坛规律、预测网络突发事件等具有一定参考价值。提出一种评价网络论坛舆论的定量指标。实验表明,该指标能有效发现网络突发事件。Studies the list of the daily amount of posters according to massive real data. Three methods are used here： the aggregated variance method, the periodogram method, and the R/S method. It is found that there are the long-range dependence processes and the self-similarity processes in the network forums. Gives the details of the estimation of the Hurst exponent. According to these experiments, presents some conclusions. First, the list of the daily amount of posters is the long-range dependence processes and the self-similarity processes. The daily amount in the fu- ture is changed as that in the past. Second, the list of the writer＇s amount of posters is the long-range dependence processes and the self-similarity processes. The amount of posters by the less poster writer is changed as that by the more poster writer. Finally, the amount of posters by the later writer is changed as that by the earlier writer in every day. These are valu- able for understanding the formation of the public opinion on Internet and forecasting the paroxysmal events. And presents a new prediction method based on the hurst exponent. The re- sult shows the exponent is the new proof to explain how the paroxysmal events happened.

关 键 词：分形 HURST指数 长相关性 自相似性 热点话题 

分 类 号：TP393.4[自动化与计算机技术—计算机应用技术]