Targeted Distillation for Sentiment Analysis
Yice Zhang, Guangyu Xie, Jingjie Lin, Jianzhu Bao, Qianlong Wang, Xi Zeng, Ruifeng Xu · 2025
Read Targeted Distillation for Sentiment Analysis by Yice Zhang, Guangyu Xie, Jingjie Lin, Jianzhu Bao, Qianlong Wang, Xi Zeng, Ruifeng Xu (2025) 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-2503-03225.txt.
Description
Yice Zhang, Ruifeng Xu, and five co-authors shrink sentiment models while keeping their skill. Their paper, from twenty twenty-five, uses targeted distillation. A big teacher model's know-how is passed to a small student. They split that target into two parts: facts to learn, and style to match.
Training runs in two stages, one per part. They also build SentiBench, a broad test set over twelve data sets. It spans many kinds of sentiment tasks, from reviews to social posts. Compact models trained their way gain a lot across tasks. They also hold up on tasks they never saw. The bench gives the field a shared yardstick.
Catalog record
| CREATOR | Yice Zhang, Guangyu Xie, Jingjie Lin, Jianzhu Bao, Qianlong Wang, Xi Zeng, Ruifeng Xu |
|---|---|
| YEAR | 2025 |
| COLLECTION | PAPERS |
| FORMAT | TXT |
| FILE SIZE | 51 KB |
| VERSION | 1.0 |
| LICENSE | http://creativecommons.org/licenses/by/4.0/ |
| RIGHTS | CC BY |
| SOURCE | arXiv |
Source: https://arxiv.org/abs/2503-03225