【作者】 李恒训； 张华平； 秦鹏； 于满泉； 刘金刚；

【Author】 LI Heng-Xun~(1),2)) ZHANG Hua-Ping~(1)) Qin Peng~(1),2)) Yu manquan~(1)) Liu Jin-Gang~(2),1)) 1)(Institution of Computing Technology,Chinese Academy of Sciences,Beijing 100190) 2)(Join Faculty of Computer Scientific Research,Capital Normal University,Beijing 100037)

【机构】 中国科学院计算技术研究所； 首都师范大学计算机联合实验室；

【摘要】 网络话题层出不穷,往往会引发重大舆情危机,如何快速高效的从海量信息中发现热点是一重大挑战。本文提出了一种基于主题词的网络热点话题发现算法。其基本思想为:首先综合主题词表和有意义串识别结果生成主题词候选集;然后对候选集进行多重过滤并采用启发式规则对主题词进行权重计算;最后,以主题词为线索,采用多特征的话题模型,融合新闻、论坛、博客的相应特征实现了网络热点话题的发现。通过在TDT4评测语料和中科院计算所天玑舆情监测系统平台上的实验分别取得了0.282的最小识别代价和93.3%的用户满意度,算法运行效率高于传统方法。实验表明,该算法对网络热点话题发现行之有效。更多还原

【Abstract】 The topic on the Internet emerges one after another,which can often trigger a major crisis in public opinion.It’s a great challenge to mine the hot spots from the vast amounts of information quickly and efficiently. This paper showed a strategy of the Internet hotspot topic detection based on keywords extraction.Its basic content can be summed up as follows:Firstly,an integrated result of thesaurus scanning and meaningful string recognition is generated for the candidate set.Then multiple filtering is used for the candidates and heuristic rules is adopted to keywords on weighting calculation.Finally,hotspot is detected using the topic model of multiple features with the keywords for clues,which integrate the corresponding feature of News,BBS and Blog.We get the min cost of 0.282 through TDT4 evaluating corpus and the satisfaction of 93.3%through the golaxy public opinion monitoring system of ICT,which is more effective than traditional method.The Experiments show that this algorithm is effective for Internet hot topic detection.更多还原