基于改进的OLDA模型话题检测及演化分析

Topic Detection and Evolution Analysis Based on Improved OLDA Model

[目的/意义]话题检测和演化分析是网络舆情监控中的热点问题,对热点话题的检测和演化分析有助于挖掘热点话题和深入理解话题的演化趋势,并给以舆情监控者提供完整的话题演化路径和更为合理的决策意见。[方法/过程]OLDA(Online Latent Dirichlet Allocation)模型是用于挖掘热点话题和分析话题演化的工具,由于其存在新旧主题混合、冗余词较多的缺点,采用双通道模式对主题、词分布的遗传度进行改进,并给出了新的词分布计算方法。[结果/结论]提出的改进OLDA模型解决了新旧主题混合问题,降低冗余词的概率,更为明确地解释话题的含义。实验表明,改进的OLDA模型更为有效地对话题进行检测及演化分析。

[ Purpose/Significance] Topic detection and evolution analysis is hot issues of network public opinion monitoring. Detection and evolution analysis of hot topics is beneficial to mine hot topics and deeply recognize the trend of topic evolution, which supplies public opinion monitoring members with the entire routes of topic evolution and more reasonable advice. [ Method/Process] OLDA（ Online La- tent Dirichlet Allocation） model is a tool used to mine hot topics and analyze evolution of topics. Considering the hybrid of old and new topics and the mass of redundant words, this paper adopts the dual channel mode with improvements of genetic degrees of theme and word distribution and puts forward a new method of calculating word distribution. [ Result/Conclusion ] Improved OLDA model figures out the hybrid of old and new topics and lowers the probability of redundant words, which will more clearly explain the meaning of the topics. The experimental results show that the improved OLDA model is more efficient to detect new topics and analyze the evolution of topics.