Large Language Models in Real-World Table Task Workflows: A Survey
Abstract: Tables are widely used across various fields such as finance, healthcare, and public administration, playing an indispensable role in modern society. Despite their importance, the structured nature of tabular data, like permutation invariance, adds complexity to its processing. Large Language Models (LLMs) offer new opportunities, but their performance remains suboptimal due to the unique characteristics of tables. Rapidly improving LLMs' ability to process tables is unattainable in the short term. Therefore, we believe that table tasks should be broken down into many interrelated subtasks to enhance performance. So, we define workflows for handling table tasks, refine existing methods based on these workflows, and compare potentially effective methods, such as LLM-based agents, for implementing all workflows, thus providing assistance for future development.
Paper Type: Long
Research Area: Machine Learning for NLP
Research Area Keywords: table QA, data-to-text generation
Contribution Types: Surveys
Languages Studied: English
Submission Number: 2634
Loading