KP20k dataset for Keyphrase Generation

About

KP20k is a dataset for benchmarking keyphrase extraction and generation models. The data is composed of 570 809 abstracts and their associated titles from scientific articles.

Details about the dataset can be found in the original paper:

Meng et al 2017. Deep keyphrase Generation Proceedings of the 55th Annual Meeting of the Association for Computational Linguistics, pages 582–592

Reference (indexer-assigned) keyphrases are also categorized under the PRMU (Present-Reordered-Mixed-Unseen) scheme as proposed in the following paper:

Florian Boudin and Ygor Gallina. 2021. Redefining Absent Keyphrases and their Effect on Retrieval Effectiveness. In Proceedings of the 2021 Conference of the North American Chapter of the Association for Computational Linguistics: Human Language Technologies, pages 4185–4193, Online. Association for Computational Linguistics.

Content

The dataset is divided into the following three splits:

Split	# documents	# keyphrases by document (average)	% Present	% Reordered	% Mixed	% Unseen
Train	530 809	5.28	40.65	7.58	24.43	27.34
Test	20 000	5.29	40.70	7.63	24.31	27.35
Validation	20 000	5.27	40.80	7.56	24.52	27.12

The following data fields are available:

id: unique identifier of the document. NB There were no ids in the original dataset. The ids were generated using the python module shortuuid (https://pypi.org/project/shortuuid/)
title: title of the document.
abstract: abstract of the document.
keyphrases: list of reference keyphrases.
prmu: list of Present-Reordered-Mixed-Unseen categories for reference keyphrases.