Question Answering with Texts and Tables through Deep Reinforcement Learning

2024-07-05

arXiv.org

AI Summary

To address the challenge of joint reasoning over textual and tabular data in open-domain multi-hop question answering, this paper proposes an end-to-end cross-modal reasoning framework based on deep reinforcement learning. Methodologically, it introduces Proximal Policy Optimization (PPO)—the first application of policy gradient methods to text-table joint QA—integrating a multimodal encoder (BERT + TabTransformer) with a differentiable table operation module to dynamically plan reading order and operational actions, eliminating the need for predefined alignments or intermediate supervision. Its key innovations include joint optimization of cross-modal reasoning paths and Monte Carlo policy evaluation. The framework achieves state-of-the-art performance on WikiTableQuestions and HybridQA, improving accuracy by 3.2% and 4.7%, respectively, demonstrating substantial gains in complex, cross-table, multi-hop reasoning capability.

Technology Category

• Natural Language Processing: Question Answering
• Search and Optimization: Learning to Search
• Reasoning under Uncertainty: Relational Probabilistic Models

Application Category

• Search and Retrieval-Augmented AI: Retrieval-Augmented Generation (RAG) and multi-modal RAG
• Semantics and Knowledge: Methods to enhance, augment, integrate or synergize semantic models such as knowledge graphs and LLMs
• Economics, Online Markets and Human Computation: Trust and reliance of crowd workers and data experts on GenAI

Problem

• Multi-hop QA with texts and tables
• Sequential model selection challenge
• Reinforcement learning for optimal tool choice

Innovation

• Deep Reinforcement Learning
• Multi-hop Answer Generation
• Open Table-and-Text Dataset