摘要
[目的/意义]如何挖掘海量学术论文中的研究主题,梳理研究主题的演化脉络和关联关系,预测主题前沿热点,对掌握科技竞争先机至关重要。[方法/过程]针对当前主题关联和预测研究中存在的不足,提出基于隐含狄利克雷(Latent Dirichlet allocation,LDA)和长短期记忆(Long Short Term Memory,LSTM)模型的研究关联与预测方法,首先基于生命周期理论划分多时序窗口,并利用LDA主题模型挖掘学术文献中的隐性研究主题,分析主题间的关联关系;基于主题预测指标的时间序列特征,运用LSTM模型对主题研究的发展趋势和研究热点进行预测,并结合基金立项和论文发表情况对预测结果进行定性修正。[结果/结论]案例分析结果表明,本文方法可以准确挖掘研究主题,分析主题关联关系,对研究主题研究走势和热点的预测具有实用价值。
[Purpose/Significance]Mining the research topics from a large number of academic literature,investigating the research evolution process and topic relation,and predicting the research fronts have significant importance to the technological competition.[Method/Process]Considering the research limitations of topic relation and prediction,a research framework of topic relation and prediction based on LDA and LSTM was proposed.First,multi-temporal windows were divided based on life cycle theory.Then,the hidden research topics from academic literature were mined based on LDA topic model,as well as the topic relation was analyzed.Furthermore,considering the characteristics of time series of topic predictors,research trends and fronts were explored based on LSTM,while the research results were improved in combination with funding programs and literature publication.[Result/Conclusion]The experimental results showed that the proposed method can accurately mine the discipline topics and analyze the topic relation,while has practical values in predicting the research trend.
作者
朱光
刘蕾
李凤景
Zhu Guang;Liu Lei;Li Fengjing(School of Management Science and Engineering,Nanjing University of Information Science andTechnology,Nanjing 210044,China)
出处
《现代情报》
CSSCI
2020年第8期38-50,共13页
Journal of Modern Information
基金
国家社会科学基金项目“泛在智慧环境下多源主体隐私感知及协同保护机制研究”(项目编号:19CTQ019)。
