Research on financial distress prediction of listed companies integrating UGC time series characteristics of stock forum

Journal Information

J. Hefei Univ. Tech. (Nat. Sci.), Mar.2025, Vol.48 No.3: 387-394

DOI: 10.3969/j.issn.1003-5060.2025.03.015

Authors

ZHANG Yu $ ^{1} $, JIANG Cuiqing $ ^{1,2} $

(1. School of Management, Hefei University of Technology, Hefei 230009, China; 2. Key Laboratory of Process Optimization and Intelligent Decision-making of Ministry of Education, Hefei 230009, China)

Abstract and Keywords

Abstract: The user generated content (UGC) of the stock forum can reflect the concerns and opinions of the stakeholders of the listed company on the company's operating performance and related events. It is timely and dynamic, and is an effective supplement to financial information. In order to effectively extract the dynamic UGC, this paper proposes a method for predicting the financial distress of listed companies that integrates the UGC time series characteristics of the stock forum. Firstly, for the time series information in user comments and user reading, considering the time series of emotional features and interactive information, the gated recurrent unit (GRU) deep recurrent network model is used to mine dynamic information in time series. Secondly, events in different time periods have different effects on financial distress prediction, and attention mechanism is used to aggregate the effects of major events on financial distress prediction. Finally, the financial distress of listed companies is predicted based on the UGC time series characteristics extracted and combined with the financial features. The research shows that the method proposed in this paper can effectively extract and aggregate time series characteristics, thus improving the prediction effect of financial distress.

Keywords: stock forum; time series characteristics; gated recurrent unit (GRU); attention mechanism; financial distress prediction

作者信息

张玉 $ ^{1} $，蒋翠清 $ ^{1,2} $

(1. 合肥工业大学管理学院，安徽合肥 230009；2. 过程优化与智能决策教育部重点实验室，安徽合肥 230009)

摘要和关键词

摘要: 股票论坛用户生成内容（user generated content，UGC）能反映上市公司利益相关者对公司经营业绩和相关事件的关注和观点，具有及时性和动态性，是对财务信息的有效补充。为有效提取动态变化 UGC，文章提出一种融入股票论坛 UGC 时序特征的上市公司财务困境预测方法。首先，针对用户评论和用户阅读中的时间序列信息，考虑情感特征时序性和互动信息时序性，采用门控循环网络（gated recurrent unit，GRU）模型，挖掘时间序列中的动态信息；其次，不同时间段下发生的事件对财务困境预测的影响程度不同，采用注意力机制聚合重大事件对财务困境预测的影响；最后，基于 UGC 时序特征，并结合财务特征对上市公司财务困境进行预测。研究表明，所提方法能够有效地提取并聚合时序特征，提高财务困境预测效果。

关键词: 股票论坛；时序特征；门控循环网络；注意力机制；财务困境预测

Funding

国家自然科学基金重点资助项目（71731005）

合肥工业大学学报(自然科学版)

Navigation