metadata
license: cc-by-4.0
A Named Entity Recognition Model for Kazakh
- The model was inspired by the LREC 2022 paper KazNERD: Kazakh Named Entity Recognition Dataset.
- The original repository for the paper can be found at https://github.com/IS2AI/KazNERD.
Differences
While the original dataset contained tokens denoting speech disfluencies and hesitations (parenthesised) and background noise [bracketed], this model was trained on a version of the dataset where such tokens were removed. Removing the tokens caused some changes in the number of sentences, tokens, and named entities (NEs).
| Dataset | Unit | Train | Valid | Test | Total |
|---|---|---|---|---|---|
| KazNERD (Original) | Sentence | 90,228 (80.06%) | 11,167 (9.91%) | 11,307 (10.03%) | 112,702 (100%) |
| KazNERD (Cleaned) | Sentence | 88,540 (80.00%) | 11,067 (10.00%) | 11,068 (10.00%) | 110,675 (100%) |
| KazNERD (Original) | Token | 1,043,305 (80.11%) | 129,223 (9.92%) | 129,824 (9.97%) | 1,302,352 (100%) |
| KazNERD (Cleaned) | Token | 1,088,461 (80.04%) | 136,021 (10.00%) | 135,426 (9.96%) | 1,359,908 (100%) |
| KazNERD (Original) | NE | 109,342 (80.20%) | 13,483 (9.89%) | 13,508 (9.91%) | 136,333 (100%) |
| KazNERD (Cleaned) | NE | 106,148 (80.17%) | 13,189 (9.96%) | 13,072 (9.87%) | 132,409 (100%) |