A Comprehensive Graph Framework for Question Answering with Mode-Seeking Preference Alignment

2025-06-22

AI Summary

Contemporary RAG-based question-answering systems suffer from insufficient global semantic understanding and generation outputs misaligned with human ethical norms and quality preferences. To address these challenges, we propose a graph-enhanced hierarchical QA framework. First, it constructs a hierarchical document graph to explicitly model cross-document semantic associations, enabling holistic contextual understanding. Second, it introduces a pattern-seeking preference optimization mechanism that aligns model outputs with human preferences via probabilistic matching constraints, emulating cognition-driven information synthesis. The method integrates retrieval-augmented generation, graph-structured modeling, semantic similarity measurement, and preference-aware optimization. Extensive experiments across six benchmark datasets demonstrate significant improvements in answer accuracy, answer consistency, and human evaluation scores. These results validate the framework’s dual advantages: deeper knowledge integration and enhanced generation controllability.

Technology Category

• Natural Language Processing: Question Answering
• Knowledge Representation and Reasoning: Preferences
• Humans and AI: Learning Human Values and Preferences

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
• Graph Algorithms and Modeling for the Web: Foundation models and LLMs for Web-related graphs

Abstract

Recent advancements in retrieval-augmented generation (RAG) have enhanced large language models in question answering by integrating external knowledge. However, challenges persist in achieving global understanding and aligning responses with human ethical and quality preferences. To address these issues, we propose GraphMPA, a comprehensive graph-based framework with mode-seeking preference alignment. Our approach constructs a hierarchical document graph using a general similarity measurement, mimicking human cognitive processes for information understanding and synthesis. Additionally, we introduce mode-seeking preference optimization to better align model outputs with human preferences through probability-matching constraints. Extensive experiments on six datasets demonstrate the effectiveness of our href{https://github.com/tangquanwei/GraphMPA}{GraphMPA}.

Problem

• Enhancing global understanding in question answering systems
• Aligning responses with human ethical and quality preferences
• Improving information synthesis via hierarchical graph structures

Innovation

• Hierarchical document graph for cognitive synthesis
• Mode-seeking preference optimization alignment
• General similarity measurement for understanding