基于Doc2Vec的期刊论文热点选题识别

Hot Topic Detection in Journal Papers Based on Doc2Vec

[目的/意义]将深度学习方法应用于热点识别的研究,实现研究热点的语义描述。[方法/过程]以教育学36本CSSCI期刊的62084篇论文为实验数据。首先,采用Doc2Vec方法对论文摘要进行向量计算;其次,对向量值进行相似度计算,生成热点选题论文集;最后,运用聚类算法和主题词提取算法获取论文热点选题的主题描述。[结果/结论]对比词频统计和共词聚类方法,本文的实验结果在研究热点的描述上具有更好的语义特征。[局限]热点选题论文集的生成受阈值的影响。深度学习可以作为揭示学科研究热点的新方法。

[Purpose/significance]In this paper,the deep learning method is applied to the research of hot topic detection to realize the semantic description of research hotspots.[Method/process] The paper selects 62084 papers from 36 education journals in the CSSCI as experimental data.Firstly,the paper carries out vector calculation by using Doc2 Vec.Secondly,the similarity calculation is performed on the vector value to generate a hot topic paper sets.Finally,the paper uses cluster analysis and keyword extraction algorithm to obtain the topic description.[Result/conclusion] Compared with word frequency statistics and co-word clustering methods,the experimental results of the paper have better semantic features in the description of research hotspots.[Limitations] The generation of hot topics is influenced by the threshold.Deep learning can be used as a new way to reveal the hotspots of academic research.