Information Extraction from Rich Text Images with RoBERTa and LION Optimizer

Localizing and extracting essential information from semi-structured business documents, such as invoices, is crucial in practical applications. This complex problem includes key information localization and extraction (KILE) and line item recognition (LIR), which depend on the choice of model and optimal training methodology. This paper presents a novel pipeline that applies RoBERTa and LION Optimizer as the primary modules for identifying and extracting crucial information on the DocILE benchmark. The experimental results indicate that the proposed method significantly improves the KILE phase with 7.24% increase in accuracy compared to the baseline and also enhances the correct recognition rate at the LIR stage.