基于HDP过程模型与学术会议的学科新兴主题发现研究 --以“人工智能”领域为例

[Purpose/Significance] In order to solve the plight of academic information overload for researchers, this article discovers emerging topics from the amounts of science and technology information, and timely tracking every field research trend, to provide auxiliary decision-making for scientific research. [Method/Process] Using Python crawler technology to acquire international conference documents of artificial intelligence from 2008 to 2017, and select no- parameters probability model (HDP) to extract topic with non-supervision mode. Then, this paper comes up with aggregative indicator named topic intensity ratio (TIR) for emerging topic discovery and combines with the novelty (NI) and conference occurrence ratio (COR) to determine topics’ stage on the life cycle (generating, emerging, mature, recession). [Result/Conclusion] This experiment used probability model (HDP) to gain 39 topics from 660 artificial intelligence documents collected, and then to discover emerging topics on the basis of the evaluation standards. The experimental results show that this method can accurately and effectively detect the emerging topics from the data source, and can make a visual analysis on the development trend of the emerging topics.