license: cc-by-nc-4.0
task_categories:
- other
language:
- sk
tags:
- keyphrase-extraction
size_categories:
- 100K<n<1M
SlovKE: A Large-Scale Dataset and LLM Evaluation for Slovak Keyphrase Extraction
This repository contains the SlovKE dataset, as presented in the paper SlovKE: A Large-Scale Dataset and LLM Evaluation for Slovak Keyphrase Extraction.
Dataset Description
SlovKE is a large-scale dataset for Slovak keyphrase extraction, consisting of 227,432 scientific abstracts with author-assigned keyphrases. The data was scraped and cleaned from the Slovak Central Register of Theses. It represents a significant resource for low-resource, morphologically rich language processing, representing a 25-fold increase over prior Slovak resources and approaching the scale of established English benchmarks like KP20K.
The dataset is designed to benchmark both unsupervised baselines (such as YAKE, TextRank, and KeyBERT) and LLM-based extraction methods (like KeyLLM). It highlights the challenges of morphological mismatch in inflected languages where surface forms in the text must be matched to canonical author-assigned keyphrases.