Leveraging NLP to Detect Governance Red Flags in Financial News and Filings

Corporate governance issues, including fraud, mismanagement, and regulatory breaches, can significantly impact market stability and investor confidence. Detecting such governance red flags early, particularly from unstructured textual sources like financial news and regulatory filings, remains a major challenge. This paper presents an NLP-based framework leveraging FinBERT embeddings and density-based clustering (HDBSCAN) to identify thematic patterns and anomalies within large-scale financial text corpora. The methodology applies UMAP for dimensionality reduction, TF-IDF for interpretability, and multiple quantitative metrics for evaluation, including silhouette scores, cosine similarity, and keyword coherence. The proposed system achieves a silhouette score of 0.83 on the NIFTY financial news dataset, demonstrating its ability to produce well-separated, semantically coherent clusters. This research provides a scalable approach for governance risk detection and contributes toward automating compliance monitoring in financial markets.