Training and Evaluating with Human Label Variation: An Empirical Study
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_ | 1866915547340341248 |
|---|---|
| author | Kurniawan, Kemal Mistica, Meladel Baldwin, Timothy Lau, Jey Han |
| author_facet | Kurniawan, Kemal Mistica, Meladel Baldwin, Timothy Lau, Jey Han |
| contents | Human label variation (HLV) challenges the standard assumption that a labelled instance has a single ground truth, instead embracing the natural variation in human annotation to train and evaluate models. While various training methods and metrics for HLV have been proposed, it is still unclear which methods and metrics perform best in what settings. We propose new evaluation metrics for HLV leveraging fuzzy set theory. Since these new proposed metrics are differentiable, we then in turn experiment with employing these metrics as training objectives. We conduct an extensive study over 6 HLV datasets testing 14 training methods and 6 evaluation metrics. We find that training on either disaggregated annotations or soft labels performs best across metrics, outperforming training using the proposed training objectives with differentiable metrics. We also show that our proposed soft micro F1 score is one of the best metrics for HLV data. |
| format | Preprint |
| id |
arxiv_https___arxiv_org_abs_2502_01891 |
| institution | arXiv |
| publishDate | 2025 |
| record_format | arxiv |
| spellingShingle | Training and Evaluating with Human Label Variation: An Empirical Study Kurniawan, Kemal Mistica, Meladel Baldwin, Timothy Lau, Jey Han Machine Learning Computation and Language Human label variation (HLV) challenges the standard assumption that a labelled instance has a single ground truth, instead embracing the natural variation in human annotation to train and evaluate models. While various training methods and metrics for HLV have been proposed, it is still unclear which methods and metrics perform best in what settings. We propose new evaluation metrics for HLV leveraging fuzzy set theory. Since these new proposed metrics are differentiable, we then in turn experiment with employing these metrics as training objectives. We conduct an extensive study over 6 HLV datasets testing 14 training methods and 6 evaluation metrics. We find that training on either disaggregated annotations or soft labels performs best across metrics, outperforming training using the proposed training objectives with differentiable metrics. We also show that our proposed soft micro F1 score is one of the best metrics for HLV data. |
| title | Training and Evaluating with Human Label Variation: An Empirical Study |
| topic | Machine Learning Computation and Language |
| url | https://arxiv.org/abs/2502.01891 |