Memory-Augmented Agent Training for Business Document Understanding
This addresses the challenge of processing specialized business documents for enterprises, particularly in logistics, by enabling LLMs to build domain expertise, though it is incremental as it adapts existing LLM agent methods with memory refinement.
The paper tackled the problem of automating transport reference extraction from business invoices, where traditional methods are manual and LLMs underperform, by introducing Matrix, a memory-augmented agent training paradigm that improved performance by 30.3% over prompting a single LLM and 35.2% over a vanilla LLM agent.
Traditional enterprises face significant challenges in processing business documents, where tasks like extracting transport references from invoices remain largely manual despite their crucial role in logistics operations. While Large Language Models offer potential automation, their direct application to specialized business domains often yields unsatisfactory results. We introduce Matrix (Memory-Augmented agent Training through Reasoning and Iterative eXploration), a novel paradigm that enables LLM agents to progressively build domain expertise through experience-driven memory refinement and iterative learning. To validate this approach, we collaborate with one of the world's largest logistics companies to create a dataset of Universal Business Language format invoice documents, focusing on the task of transport reference extraction. Experiments demonstrate that Matrix outperforms prompting a single LLM by 30.3%, vanilla LLM agent by 35.2%. We further analyze the metrics of the optimized systems and observe that the agent system requires less API calls, fewer costs and can analyze longer documents on average. Our methods establish a new approach to transform general-purpose LLMs into specialized business tools through systematic memory enhancement in document processing tasks.