File size: 1,476 Bytes
71a3e71 35739cc 71a3e71 2d9ff1a 91bc38d 35739cc 91bc38d 35739cc |
1 2 3 4 5 6 7 8 9 10 11 12 13 14 15 16 17 18 19 20 21 22 23 24 25 26 27 28 29 30 31 32 33 34 35 36 37 38 39 40 41 42 43 44 45 46 47 48 49 50 51 52 53 54 55 56 57 58 59 |
---
license: cc
task_categories:
- text-generation
- text2text-generation
language:
- en
tags:
- keyphrase-generation
- domain-adaptation
- paleontology
- astrophysics
- natural-language-processing
size_categories:
- 1K<n<10K
configs:
- config_name: paleo
data_files:
- split: train
path:
- "paleo/train.jsonl"
- split: test
path: "paleo/test.jsonl"
- config_name: nlp
data_files:
- split: train
path:
- "nlp/train.jsonl"
- split: test
path: "nlp/test.jsonl"
- config_name: astro
data_files:
- split: train
path:
- "astro/train.jsonl"
- split: test
path: "astro/test.jsonl"
---
# `silk` synthetic training samples and human-labeled test sets for domain adaptation in keyphrase generation
This dataset contains the synthetic samples generated by 🧵 `silk`, a method that leverages citation contexts to create synthetic samples of documents paired with silver-standard keyphrases for adapting keyphrase generation models to new domains.
We applied `silk` on three domains: Natural Language Processing (nlp), Astrophysics (astro) and Paleontology (paleo).
This dataset also includes three human-labeled test sets to assess the performance of keyphrase generation across these domains.
## Citation
If you use this dataset, please cite the following paper:
```
Florian Boudin and Akiko Aizawa.
Unsupervised Domain Adaptation for Keyphrase Generation using Citation Context,
Proceedings of EMNLP 2024 (Findings).
```
|