Constraint Multi-class Positive and Unlabeled Learning for Distantly Supervised Named Entity Recognition
Yuzhe Zhang, Min Cen, Hong Zhang · 2025
Read Constraint Multi-class Positive and Unlabeled Learning for Distantly Supervised Named Entity Recognition by Yuzhe Zhang, Min Cen, Hong Zhang (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-2504-04963.txt.
Description
Yuzhe Zhang, Min Cen, and Hong Zhang wrote this 2025 arXiv paper on teaching computers to find names in text without hand-labeled training data. Distant supervision labels text automatically from knowledge bases, but those bases are incomplete, so many real names go unlabeled and the learner is punished for finding them. The authors recast the problem as learning from positives and unlabeled examples across several name classes at once.
Their method adds a constraint factor to the risk estimate that earlier approaches use, which they show makes the learner harder to overfit when labeled positives are scarce. The theory is worked out formally, not just asserted. On two benchmark data sets labeled by different knowledge sources, the method beat existing distantly supervised systems. The audience is anyone building entity extraction where annotation budgets are zero and knowledge bases are patchy.
Catalog record
| CREATOR | Yuzhe Zhang, Min Cen, Hong Zhang |
|---|---|
| YEAR | 2025 |
| COLLECTION | PAPERS |
| FORMAT | TXT |
| FILE SIZE | 36 KB |
| VERSION | 1.0 |
| LICENSE | http://creativecommons.org/licenses/by/4.0/ |
| RIGHTS | CC BY |
| SOURCE | arXiv |
Source: https://arxiv.org/abs/2504-04963