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  

Detecting Emerging Topics of Subjects Based on HDP Model and Academic Conferences: A Case Study in the Field of Artificial Intelligence

作　　者：杨金庆 肖兵 程秀峰[1] 叶光辉[1] Yang Jinqing

机构地区：[1]华中师范大学信息管理学院,湖北武汉430079

出　　处：《情报理论与实践》2019年第4期117-122,共6页Information Studies:Theory & Application

基　　金：国家社会科学基金重大项目"基于多维度聚合的网络资源知识发现研究"(项目编号:13&ZD183);国家自然科学基金青年科学基金项目"基于QSIM的图书馆移动用户群体行为模拟与学习兴趣引导研究"(项目编号:71503097);中央高校基本科研业务费项目"基于社会化标签挖掘的城市画像研究"(项目编号:CCNU18QN040);研究生自主科研立项研究的成果之一

摘　　要：[目的/意义]针对研究人员所面临的学术信息过载的困境,尝试从海量科技会议的议题报告中发现新兴主题,及时跟踪各领域研究动向,为科研人员的科学研究提供辅助决策。[方法/过程]首先采集某领域国际会议的议题报告,然后通过无参数概率主题模型(HDP)对文档进行无监督主题抽取,进而结合新颖度(NI)、会议出现率(COR)以及主题强度比(TIR)等指标,对主题所处的生命周期(生成、新兴、成熟、衰老)阶段进行判定。[结果/结论]利用上述方法对2008—2017年人工智能领域660份议题报告的分析,共抽取39个主题,对主题所处阶段进行判定,从而发现目标新兴主题。结果表明:该方法能准确高效地识别新兴主题,同时能为新兴主题趋势预测提供可视化。[Purpose/significance] In order to solve the plight of academic information overload for researchers,this paper tries to detect emerging topics from massive scientific and technical conferences,and timely track research dynamics in different fields,providing auxiliary decision-making for scientific research.[Method/process] First,the paper acquires international conference topics of a specific field,and extracts topics from the documents with non-supervision mode through hierarchical dirichlet process(HDP) model.Then,the paper determines topics’ stage in the life cycle(generation,emergence,mature,recession) by the indicators such as novelty(NI),conference occurrence ratio(COR),and topic intensity ratio(TIR).[Result/conclusion] This experiment extracts 39 topics from 660 topic reports in the field of artificial intelligence from 2008 to 2017 through HDP model,determines the life stage of these topics,and then discovers target emerging topics.The experimental results show that the proposed method can accurately and effectively detect emerging topics,and at the same time make a visual analysis on the development trend of emerging topics.

关 键 词：新兴主题 学术会议 HDP模型 人工智能 

分 类 号：G254[文化科学—图书馆学] TP18[自动化与计算机技术—控制理论与控制工程]