DocLLM: A layout-aware generative language model for multimodal document understanding

Meta info.

• Authors: Dongsheng Wang, Natraj Raman, Mathieu Sibue et al.
• Paper: https://arxiv.org/pdf/2401.00908.pdf
• Affiliation: JP Morgan AI Research

TL; DR

multi-modal LLM에서 착안, LM이 text와 (정형화된 document 내에서) 위치정보를 input으로 받도록 하여 internal structured document understanding 문제 해결