ReFeed: Multi-dimensional Summarization Refinement with Reflective Reasoning on Feedback

2025-03-27

AI Summary

Multidimensional summary refinement faces challenges including difficulty in balancing multiple dimensions and inefficient utilization of feedback. This paper proposes the first end-to-end refinement framework based on reflective reasoning, leveraging Chain-of-Thought (CoT) reasoning over extended sequences (Long-CoT) to model couplings among multidimensional feedback signals, thereby jointly optimizing coherence, factual consistency, and conciseness. Key contributions include: (1) a novel reflective reasoning mechanism that explicitly models the interplay among feedback exposure volume, dimension count, and reasoning strategy; (2) SumFeed-CoT—the first Long-CoT–based dataset for reflective reasoning optimization in summarization feedback; and (3) a lightweight architecture with a noise-robust, structured feedback fusion method. Experiments show that our model achieves state-of-the-art performance despite a 40% reduction in parameter count, yielding ROUGE improvements of 3.2–5.8 points across multiple benchmarks, while demonstrating strong robustness to feedback noise and ordering variations.

Technology Category

• Natural Language Processing: Summarization
• Machine Learning: Large Multimodal Models (LMMs)
• Search and Optimization: Metareasoning and Metaheuristics

Application Category

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

Abstract

Summarization refinement faces challenges when extending to multi-dimension. In this paper, we introduce ReFeed, a powerful summarization refinement pipeline that enhances multiple dimensions through reflective reasoning on feedback. To achieve this, we release SumFeed-CoT, a large-scale Long-CoT-based dataset optimized for training a lightweight model with reflective reasoning. Our experiments reveal how the number of dimensions, feedback exposure, and reasoning policy influence refinement performance, highlighting reflective reasoning and simultaneously addressing multiple feedback is crucial to mitigate trade-off between dimensions. Furthermore, ReFeed is robust to noisy feedback and feedback order. Lastly, our finding emphasizes that creating data with a proper goal and guideline constitutes a fundamental pillar of effective reasoning. The dataset and model will be released.

Problem

• Enhancing multi-dimensional summarization refinement via reflective reasoning
• Addressing trade-offs between dimensions through simultaneous feedback processing
• Developing robust refinement models resistant to noisy and unordered feedback

Innovation

• ReFeed pipeline enhances multi-dimensional summarization refinement
• SumFeed-CoT dataset trains lightweight model with reflective reasoning
• Robust to noisy feedback and varying feedback order