Application of News Text Mining in Urban Credit Assessment: A Study Based on BERT and QDA-DTM Models

Abstract

This paper focuses on news text mining and urban credit assessment, developing a dual-dimensional hybrid algorithm model that integrates BERT sentiment analysis with QDA-DTM thematic modeling. The study first obtained a large amount of news text data from multiple provinces across China via the “Credit China” platform. After preprocessing, including domain-adaptive word segmentation, the BERT model was used for sentiment analysis. By cross-validating sentiment and semantic features, the reliability of credit risk monitoring was enhanced. Additionally, the QDA model is employed to optimize the spatio-temporal distribution characteristics of themes, while the DTM model is used to capture the long-term dynamic evolution patterns of themes. Experimental results demonstrate that this dual-dimensional algorithmic model outperforms traditional methods in credit event prediction, providing more reliable support for credit risk early warning.