ReFeed: Multi-dimensional Summarization Refinement with Reflective Reasoning on Feedback
Fuente:
arXiv
Gespeichert in:
| Hauptverfasser: | , , , , , , |
|---|---|
| Format: | Preprint |
| Veröffentlicht: |
2025
|
| Schlagworte: | |
| Online-Zugang: | |
| Tags: |
Tag hinzufügen
Keine Tags, Fügen Sie den ersten Tag hinzu!
|
| _version_ | 1866910896260907008 |
|---|---|
| author | Yun, Taewon Oh, Jihwan Min, Hyangsuk Lee, Yuho Bang, Jihwan Cai, Jason Song, Hwanjun |
| author_facet | Yun, Taewon Oh, Jihwan Min, Hyangsuk Lee, Yuho Bang, Jihwan Cai, Jason Song, Hwanjun |
| contents | 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. |
| format | Preprint |
| id |
arxiv_https___arxiv_org_abs_2503_21332 |
| institution | arXiv |
| publishDate | 2025 |
| record_format | arxiv |
| spellingShingle | ReFeed: Multi-dimensional Summarization Refinement with Reflective Reasoning on Feedback Yun, Taewon Oh, Jihwan Min, Hyangsuk Lee, Yuho Bang, Jihwan Cai, Jason Song, Hwanjun Computation and Language Artificial Intelligence 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. |
| title | ReFeed: Multi-dimensional Summarization Refinement with Reflective Reasoning on Feedback |
| topic | Computation and Language Artificial Intelligence |
| url | https://arxiv.org/abs/2503.21332 |