基于BERT-LDA的国外LIS领域学科交叉研究演化分析与前沿主题识别

Evolutionary Analysis and Frontier Identification of Interdisciplinary Research in Foreign LIS Field Based on BERT-LDA

识别学科交叉研究的前沿主题,并对演化趋势进行分析,有助于揭示学科交叉融合的方向,为未来创新性、突破性研究提供参考。首先,基于引文视角构建测度论文学科交叉性的指标,识别具有学科交叉性的研究论文;其次,通过BERT-LDA模型识别研究主题,利用余弦相似度计算主题之间的相似度,构建主题演化路径;最后,基于新颖度、增长性、关注度、影响力构建前沿主题识别指标体系,识别具有前沿性的学科交叉研究主题。以图书情报学(Library and Information Science,LIS)为例展开研究,研究结果显示,2004—2023年该学科领域的交叉研究主题呈现出逐渐细化和深入的特点,主要集中在信息挖掘与知识发现、互联网信息行为、医疗信息学3个方面;现阶段学科交叉研究前沿主题为医疗数据模型、舆情治理与情感分析、机器学习与深度学习;基于信息技术的研究方法和其在不同领域的应用研究具有良好的应用前景,有可能成为未来LIS领域的核心研究主题。

Identifying interdisciplinary research frontier topics and analyzing evolutionary trends can help reveal the direction of interdisciplinary integration,and provide references for future innovative and breakthrough research.Firstly,based on the perspective of citation,we construct indicators to measure the interdisciplinary nature of research papers,and identify research papers with interdisciplinary nature.Secondly,we use the BERT-LDA model to identify research topics,calculate the similarity between topics using cosine similarity,and construct topic evolution paths.Finally,we construct a research frontier topic identification index system that includes topic novelty,topic growth,topic attention,and topic influence to identify interdisciplinary research topics with cutting-edge characteristics.This paper takes the Library and Information Science(LIS)discipline as an example to conduct research.The results show that interdisciplinary research topics in this field from 2004 to 2023 have gradually become more refined and in-depth,mainly focusing on information mining and knowledge discovery,Internet information behavior,and medical informatics.At present,the interdisciplinary topics of medical data model,public opinion governance and sentiment analysis,and machine learning and deep learning have cutting-edge characteristics.Research methods based on information technology and their applications in different fields have good application prospects and may become the core research topics in the future field of LIS.