IndoPref: A Multi-Domain Pairwise Preference Dataset for Indonesian
Fuente:
arXiv
Gespeichert in:
| Hauptverfasser: | , , , |
|---|---|
| Format: | Preprint |
| Veröffentlicht: |
2025
|
| Schlagworte: | |
| Online-Zugang: | |
| Tags: |
Tag hinzufügen
Keine Tags, Fügen Sie den ersten Tag hinzu!
|
| _version_ | 1866914152320073728 |
|---|---|
| author | Wiyono, Vanessa Rebecca Anugraha, David Purwarianti, Ayu Winata, Genta Indra |
| author_facet | Wiyono, Vanessa Rebecca Anugraha, David Purwarianti, Ayu Winata, Genta Indra |
| contents | Over 200 million people speak Indonesian, yet the language remains significantly underrepresented in preference-based research for large language models (LLMs). Most existing multilingual datasets are derived from English translations, often resulting in content that lacks cultural and linguistic authenticity. To address this gap, we introduce IndoPref, the first fully human-authored and multi-domain Indonesian preference dataset designed to evaluate the naturalness and quality of LLM-generated text. The dataset contains 522 prompts and yields 4,099 human-annotated pairwise preferences from comparisons across five instruction-tuned LLMs. All annotations are natively written in Indonesian with strong inter-annotator agreement, measured by Krippendorff's alpha. Our benchmark spans 10 diverse categories, enabling practitioners to identify LLMs' fine-grained strengths and weaknesses. |
| format | Preprint |
| id |
arxiv_https___arxiv_org_abs_2507_22159 |
| institution | arXiv |
| publishDate | 2025 |
| record_format | arxiv |
| spellingShingle | IndoPref: A Multi-Domain Pairwise Preference Dataset for Indonesian Wiyono, Vanessa Rebecca Anugraha, David Purwarianti, Ayu Winata, Genta Indra Computation and Language Artificial Intelligence Over 200 million people speak Indonesian, yet the language remains significantly underrepresented in preference-based research for large language models (LLMs). Most existing multilingual datasets are derived from English translations, often resulting in content that lacks cultural and linguistic authenticity. To address this gap, we introduce IndoPref, the first fully human-authored and multi-domain Indonesian preference dataset designed to evaluate the naturalness and quality of LLM-generated text. The dataset contains 522 prompts and yields 4,099 human-annotated pairwise preferences from comparisons across five instruction-tuned LLMs. All annotations are natively written in Indonesian with strong inter-annotator agreement, measured by Krippendorff's alpha. Our benchmark spans 10 diverse categories, enabling practitioners to identify LLMs' fine-grained strengths and weaknesses. |
| title | IndoPref: A Multi-Domain Pairwise Preference Dataset for Indonesian |
| topic | Computation and Language Artificial Intelligence |
| url | https://arxiv.org/abs/2507.22159 |