REMOAC: A Retroactive Explainable Method for OCR Anomalies Correction in Legal Domain

Efficient OCR Anomaly Detection and Correction is essential in the legal domain, as it significantly enhances the ability of legal professionals to extract accurate information from documents. This paper presents a novel approach called REMOAC, that improves the performance of a legal text classifier through OCR anomaly detection and correction in legal documents, by actively using state-of-the-art models and explainability techniques. Explainability is a key aspect of our approach, both because we provide transparency and comprehensibility in the OCR anomaly correction process and, more importantly, because we actively use it to improve classification performance.