DocGraphLM: Documental Graph Language Model for Information Extraction
This work addresses information extraction for documents with complex layouts, representing an incremental advancement by integrating graph features into existing architectures.
The paper tackles information extraction from visually rich documents by introducing DocGraphLM, a framework that combines pre-trained language models with graph semantics, resulting in consistent improvements on IE and QA tasks across three SotA datasets and accelerated convergence during training.
Advances in Visually Rich Document Understanding (VrDU) have enabled information extraction and question answering over documents with complex layouts. Two tropes of architectures have emerged -- transformer-based models inspired by LLMs, and Graph Neural Networks. In this paper, we introduce DocGraphLM, a novel framework that combines pre-trained language models with graph semantics. To achieve this, we propose 1) a joint encoder architecture to represent documents, and 2) a novel link prediction approach to reconstruct document graphs. DocGraphLM predicts both directions and distances between nodes using a convergent joint loss function that prioritizes neighborhood restoration and downweighs distant node detection. Our experiments on three SotA datasets show consistent improvement on IE and QA tasks with the adoption of graph features. Moreover, we report that adopting the graph features accelerates convergence in the learning process during training, despite being solely constructed through link prediction.