DocReward: A Document Reward Model for Structuring and Stylizing
Fuente:
arXiv
Salvato in:
| Autori principali: | , , , , , , , , , , , , , , , , , , , |
|---|---|
| Natura: | Preprint |
| Pubblicazione: |
2025
|
| Soggetti: | |
| Accesso online: | |
| Tags: |
Aggiungi Tag
Nessun Tag, puoi essere il primo ad aggiungerne!!
|
| _version_ | 1866913138878709760 |
|---|---|
| author | Liu, Junpeng Zhao, Yuzhong Cao, Bowen Ding, Jiayu Jia, Yilin Lv, Tengchao Huang, Yupan Wu, Wenshan Huang, Shaohan Yang, Nan Dong, Li Cui, Lei Ge, Tao Wang, Xun Jiao, Huitian Mao, Sun Kartik, FNU Chen, Si-Qing Lam, Wai Wei, Furu |
| author_facet | Liu, Junpeng Zhao, Yuzhong Cao, Bowen Ding, Jiayu Jia, Yilin Lv, Tengchao Huang, Yupan Wu, Wenshan Huang, Shaohan Yang, Nan Dong, Li Cui, Lei Ge, Tao Wang, Xun Jiao, Huitian Mao, Sun Kartik, FNU Chen, Si-Qing Lam, Wai Wei, Furu |
| contents | Recent agentic workflows automate professional document generation but focus narrowly on textual quality, overlooking structural and stylistic professionalism, which is equally critical for readability. This gap stems mainly from a lack of effective reward models capable of guiding agents toward producing documents with high structural and stylistic professionalism. We introduce DocReward, a document reward model that evaluates documents based on their structure and style. To achieve this, we propose a textual-quality-agnostic framework that ensures assessments are not confounded by content quality, and construct DocPair, a dataset of 117K paired documents covering 32 domains and 267 types. Each pair shares identical content but differs in structural and stylistic professionalism. DocReward is trained using the Bradley-Terry loss. On a manually annotated benchmark, DocReward outperforms GPT-5 by 14.6 percentage points in the same setting. Reinforcement learning experiments further show that DocReward effectively guides agents toward generating documents with consistently higher structural and stylistic professionalism, highlighting its practical utility. |
| format | Preprint |
| id |
arxiv_https___arxiv_org_abs_2510_11391 |
| institution | arXiv |
| publishDate | 2025 |
| record_format | arxiv |
| spellingShingle | DocReward: A Document Reward Model for Structuring and Stylizing Liu, Junpeng Zhao, Yuzhong Cao, Bowen Ding, Jiayu Jia, Yilin Lv, Tengchao Huang, Yupan Wu, Wenshan Huang, Shaohan Yang, Nan Dong, Li Cui, Lei Ge, Tao Wang, Xun Jiao, Huitian Mao, Sun Kartik, FNU Chen, Si-Qing Lam, Wai Wei, Furu Computer Vision and Pattern Recognition Artificial Intelligence Computation and Language Recent agentic workflows automate professional document generation but focus narrowly on textual quality, overlooking structural and stylistic professionalism, which is equally critical for readability. This gap stems mainly from a lack of effective reward models capable of guiding agents toward producing documents with high structural and stylistic professionalism. We introduce DocReward, a document reward model that evaluates documents based on their structure and style. To achieve this, we propose a textual-quality-agnostic framework that ensures assessments are not confounded by content quality, and construct DocPair, a dataset of 117K paired documents covering 32 domains and 267 types. Each pair shares identical content but differs in structural and stylistic professionalism. DocReward is trained using the Bradley-Terry loss. On a manually annotated benchmark, DocReward outperforms GPT-5 by 14.6 percentage points in the same setting. Reinforcement learning experiments further show that DocReward effectively guides agents toward generating documents with consistently higher structural and stylistic professionalism, highlighting its practical utility. |
| title | DocReward: A Document Reward Model for Structuring and Stylizing |
| topic | Computer Vision and Pattern Recognition Artificial Intelligence Computation and Language |
| url | https://arxiv.org/abs/2510.11391 |