AGSC: Adaptive Granularity and Semantic Clustering for Uncertainty Quantification in Long-text Generation
Guanran Luo, Wentao Qiu, Wanru Zhao, Wenhan Lv, Zhongquan Jian Meihong Wang , Qingqiang Wu School of Informatics, Xiamen University [email protected], [email protected] · 2026
Read AGSC: Adaptive Granularity and Semantic Clustering for Uncertainty Quantification in Long-text Generation by Guanran Luo, Wentao Qiu, Wanru Zhao, Wenhan Lv, Zhongquan Jian Meihong Wang , Qingqiang Wu School of Informatics, Xiamen University [email protected], [email protected] (2026) in the Tower of Babel, a free online library of public-domain works preserved in clean, searchable text. No sign-up. The full text is served as a plain-text file at a predictable URL, where text is present: https://antarctic-labs.com/tower-of-babel/arxiv-2604-06812.txt.
Description
Luo and six colleagues quantify uncertainty in long-form generation with AGSC. Hallucination haunts long outputs, yet reliable uncertainty scores struggle across mixed themes, and fine-grained splitting costs heavily while missing neutral information. AGSC first uses NLI neutral probabilities as triggers to separate irrelevance from uncertainty, cutting wasted compute. Gaussian mixture soft clustering then models latent themes and weights topics for aggregation. Tests on BIO and LongFact show leading correlation with factuality while cutting inference time by about 60 percent versus full atomic splitting.
Catalog record
| CREATOR | Guanran Luo, Wentao Qiu, Wanru Zhao, Wenhan Lv, Zhongquan Jian Meihong Wang , Qingqiang Wu School of Informatics, Xiamen University [email protected], [email protected] |
|---|---|
| YEAR | 2026 |
| COLLECTION | PAPERS |
| FORMAT | TXT |
| FILE SIZE | 43455 |
| LICENSE | http://creativecommons.org/licenses/by/4.0/ |
| RIGHTS | CC BY |
| SOURCE | arXiv |
Source: https://arxiv.org/abs/2604.06812