Dynamic Attention-Guided Context Decoding for Mitigating Context Faithfulness Hallucinations in Large Language Models

2025-01-02

AI Summary

To address context faithfulness hallucinations in large language models—caused by insufficient context utilization and high output uncertainty—this paper proposes a lightweight, single-pass decoding framework with dynamic attention guidance. The method innovatively jointly models attention distributions and uncertainty signals; probe-based analysis empirically validates that attention strength correlates with context utilization, enabling dynamic decoding control. It integrates attention mechanism analysis, uncertainty estimation, context-aware decoding, and lightweight post-processing. Evaluated on multi-source question-answering benchmarks, the approach reduces hallucination rates by 27.3% on average, significantly improving output faithfulness and robustness while maintaining low computational overhead.

Technology Category

• Machine Learning: Calibration & Uncertainty Quantification
• Reasoning under Uncertainty: Relational Probabilistic Models
• Natural Language Processing: Safety and Robustness

Application Category

• User Modeling, Personalization and Recommendation: Fairness-aware retrieval and ranking
• Search and Retrieval-Augmented AI: Web learning to rank, online learning, and counterfactual learning for ranking
• Semantics and Knowledge: Methods to enhance, augment, integrate or synergize semantic models such as knowledge graphs and LLMs

Abstract

Large language models (LLMs) often suffer from context faithfulness hallucinations, where outputs deviate from retrieved information due to insufficient context utilization and high output uncertainty. Our uncertainty evaluation experiments reveal a strong correlation between high uncertainty and hallucinations. We hypothesize that attention mechanisms encode signals indicative of contextual utilization, validated through probing analysis. Based on these insights, we propose Dynamic Attention-Guided Context Decoding (DAGCD), a lightweight framework that integrates attention distributions and uncertainty signals in a single-pass decoding process. Experiments across QA datasets demonstrate DAGCD's effectiveness, achieving significant improvements in faithfulness and robustness while maintaining computational efficiency.

Problem

• Large Language Models
• Accuracy Issues
• Context Understanding

Innovation

• DAGCD
• Attention Mechanism
• Uncertainty Signal Integration