论文部分内容阅读
【目的】对基于主题模型的演化方法进行梳理与分析,总结各方法优缺点及在情报分析领域的适用性。【文献范围】从Google Scholar、Web of Science中以“Topic/Theme Evolution”、“Time Topic Model”、“Dynamic Topic Model”为关键词/主题词进行文献检索,结合引文查询,经阅读后筛选出25篇作为本文的参考文献。【方法】采用文献分析法,对比各模型实现机制与功能特征,总结不同种类模型的优缺点及适用领域。【结果】目前的主题演化模型主要在可变主题数、支持在线分析、连续时间窗三个维度进行实现,大多数系统具备1—2个功能,基本可以满足情报分析的应用需求。【局限】对一些模型的具体实现分析不够深入。【结论】不同来源、不同粒度、不同时间窗的演化分析应该针对具体应用需求,结合模型特点使用相应的主题模型演化方法。
【Objective】 Combining and analyzing the evolutionary methods based on thematic models, the author summarizes the advantages and disadvantages of each method and its applicability in the field of intelligence analysis. 【Scope】 Literature search is carried out from Google Scholar and Web of Science by keyword / keyword with “Topic / Theme Evolution”, “Time Topic Model” and “Dynamic Topic Model” After screening, 25 articles were selected as references in this article. 【Method】 By means of literature analysis, the mechanism and function characteristics of each model were compared with each other, and the advantages and disadvantages of different kinds of models and their applicable fields were summarized. 【Result】 The current theme evolution model is mainly implemented in three dimensions: variable topics, online analysis support and continuous time window. Most systems have 1-2 functions, which can basically meet the application requirements of intelligence analysis. Limitations The analysis of the concrete implementation of some models is not deep enough. 【Conclusion】 The evolution analysis of different sources, different granularities and different time windows should be based on the specific application requirements, and the corresponding theme model evolution method should be used in combination with the characteristics of the model.