Publication detail

Query-Based Keyphrase Extraction from Long Documents

DOČEKAL, M. SMRŽ, P.

Original Title

Query-Based Keyphrase Extraction from Long Documents

Type

conference paper

Language

English

Original Abstract

Transformer-based architectures in natural language processing force input size limits that can be problematic when long documents need to be processed. This paper overcomes this issue for keyphrase extraction by chunking the long documents while keeping a global context as a query defining the topic for which relevant keyphrases should be extracted. The developed system employs a pre-trained BERT model and adapts it to estimate the probability that a given text span forms a keyphrase. We experimented using various context sizes on two popular datasets, Inspec and SemEval, and a large novel dataset. The presented results show that a shorter context with a query overcomes a longer one without the query on long documents.

Keywords

keyphrase,keyword,long documents,query-based keyphrase extraction,BERT,transformer

Authors

DOČEKAL, M.; SMRŽ, P.

Released

4. 5. 2022

Publisher

LibraryPress@UF

Location

Jensen Beach

ISBN

2334-0762

Year of study

2022

Number

35

Pages from

1

Pages to

4

Pages count

4

URL

https://journals.flvc.org/FLAIRS/article/view/130737

BibTex

@inproceedings{BUT179282,
  author="Martin {Dočekal} and Pavel {Smrž}",
  title="Query-Based Keyphrase Extraction from Long Documents",
  booktitle="The International FLAIRS Conference Proceedings",
  year="2022",
  series="2022",
  volume="2022",
  number="35",
  pages="1--4",
  publisher="LibraryPress@UF",
  address="Jensen Beach",
  doi="10.32473/flairs.v35i.130737",
  issn="2334-0762",
  url="https://journals.flvc.org/FLAIRS/article/view/130737"
}