【摘要】 [目的/意义]针对现有企业弱信号识别方法的单一性,本文提出一种基于LDA-BERT融合模型的弱信号全自动识别方法。[方法/过程]该方法首先通过LDA主题模型对文本数据集进行分类;其次,构建主题和术语双层过滤函数从主题分类的结果中提取早期预警信号,通过紧密中心度、主题权重以及主题自相关性三大度量对主题进行过滤,并基于主题内术语的归一化频率和概率提取出弱信号;最后,运用BERT深度学习模型在语义上拓展弱信号。[结果/结论]本文使用企业社交媒体新闻数据集对构建的系统模型进行验证,有效检测出相关弱信号,并挖掘出弱信号随时间推移逐渐增强的演化特性。该模型不仅解决了现有弱信号研究人工参与较多和检测结果可解释能力不高的问题,且融合模型弥补了LDA词袋模型的不足,能更有效地对弱信号进行识别,为企业危机预警和战略决策管理提供参考信息的同时,也为弱信号识别研究提供了新思路、新方法。更多还原

【Abstract】 [Purpose/Significance] Aiming at the singularity of the existing weak signal identification methods for enterprises,this paper proposes an automatic weak signal identification method based on the LDA-BERT fusion model.[Method/Process] This method first classifies the text data set through the LDA topic model,and secondly,constructs a two-layer filter function for topics and terms to extract early warning signals from the results of topic classification,and uses three major metrics: close centrality,topic weight,and topic autocorrelation to filter the subject,and weak signals are extracted based on the normalized frequency and probability of the terms in the subject. Finally,use the BERT deep learning model to semantically expand weak signals. [Result/Conclusion] This article uses the corporate social media news data set to verify the built system model,effectively detecting related weak signals,and digging out the evolution characteristics of weak signals that gradually increase over time. This model not only solves the problems of the existing weak signal research with more manual participation and low interpretability of the detection results,but also use the fusion model to make up the lack of the LDA bag-of-words model,making it more effective in identifying weak signals. While providing reference information for corporate crisis early warning and strategic decision management,it also provides new ideas and new methods for weak signal identification research.更多还原