@inproceedings{golchin-etal-2023-mask,
title = "Do not Mask Randomly: Effective Domain-adaptive Pre-training by Masking In-domain Keywords",
author = "Golchin, Shahriar and
Surdeanu, Mihai and
Tavabi, Nazgol and
Kiapour, Ata",
editor = "Can, Burcu and
Mozes, Maximilian and
Cahyawijaya, Samuel and
Saphra, Naomi and
Kassner, Nora and
Ravfogel, Shauli and
Ravichander, Abhilasha and
Zhao, Chen and
Augenstein, Isabelle and
Rogers, Anna and
Cho, Kyunghyun and
Grefenstette, Edward and
Voita, Lena",
booktitle = "Proceedings of the 8th Workshop on Representation Learning for NLP (RepL4NLP 2023)",
month = jul,
year = "2023",
address = "Toronto, Canada",
publisher = "Association for Computational Linguistics",
url = "https://aclanthology.org/2023.repl4nlp-1.2",
doi = "10.18653/v1/2023.repl4nlp-1.2",
pages = "13--21",
abstract = "We propose a novel task-agnostic in-domain pre-training method that sits between generic pre-training and fine-tuning. Our approach selectively masks in-domain keywords, i.e., words that provide a compact representation of the target domain. We identify such keywords using KeyBERT (Grootendorst, 2020). We evaluate our approach using six different settings: three datasets combined with two distinct pre-trained language models (PLMs). Our results reveal that the fine-tuned PLMs adapted using our in-domain pre-training strategy outperform PLMs that used in-domain pre-training with random masking as well as those that followed the common pre-train-then-fine-tune paradigm. Further, the overhead of identifying in-domain keywords is reasonable, e.g., 7-15{\%} of the pre-training time (for two epochs) for BERT Large (Devlin et al., 2019).",
}
<?xml version="1.0" encoding="UTF-8"?>
<modsCollection xmlns="http://www.loc.gov/mods/v3">
<mods ID="golchin-etal-2023-mask">
<titleInfo>
<title>Do not Mask Randomly: Effective Domain-adaptive Pre-training by Masking In-domain Keywords</title>
</titleInfo>
<name type="personal">
<namePart type="given">Shahriar</namePart>
<namePart type="family">Golchin</namePart>
<role>
<roleTerm authority="marcrelator" type="text">author</roleTerm>
</role>
</name>
<name type="personal">
<namePart type="given">Mihai</namePart>
<namePart type="family">Surdeanu</namePart>
<role>
<roleTerm authority="marcrelator" type="text">author</roleTerm>
</role>
</name>
<name type="personal">
<namePart type="given">Nazgol</namePart>
<namePart type="family">Tavabi</namePart>
<role>
<roleTerm authority="marcrelator" type="text">author</roleTerm>
</role>
</name>
<name type="personal">
<namePart type="given">Ata</namePart>
<namePart type="family">Kiapour</namePart>
<role>
<roleTerm authority="marcrelator" type="text">author</roleTerm>
</role>
</name>
<originInfo>
<dateIssued>2023-07</dateIssued>
</originInfo>
<typeOfResource>text</typeOfResource>
<relatedItem type="host">
<titleInfo>
<title>Proceedings of the 8th Workshop on Representation Learning for NLP (RepL4NLP 2023)</title>
</titleInfo>
<name type="personal">
<namePart type="given">Burcu</namePart>
<namePart type="family">Can</namePart>
<role>
<roleTerm authority="marcrelator" type="text">editor</roleTerm>
</role>
</name>
<name type="personal">
<namePart type="given">Maximilian</namePart>
<namePart type="family">Mozes</namePart>
<role>
<roleTerm authority="marcrelator" type="text">editor</roleTerm>
</role>
</name>
<name type="personal">
<namePart type="given">Samuel</namePart>
<namePart type="family">Cahyawijaya</namePart>
<role>
<roleTerm authority="marcrelator" type="text">editor</roleTerm>
</role>
</name>
<name type="personal">
<namePart type="given">Naomi</namePart>
<namePart type="family">Saphra</namePart>
<role>
<roleTerm authority="marcrelator" type="text">editor</roleTerm>
</role>
</name>
<name type="personal">
<namePart type="given">Nora</namePart>
<namePart type="family">Kassner</namePart>
<role>
<roleTerm authority="marcrelator" type="text">editor</roleTerm>
</role>
</name>
<name type="personal">
<namePart type="given">Shauli</namePart>
<namePart type="family">Ravfogel</namePart>
<role>
<roleTerm authority="marcrelator" type="text">editor</roleTerm>
</role>
</name>
<name type="personal">
<namePart type="given">Abhilasha</namePart>
<namePart type="family">Ravichander</namePart>
<role>
<roleTerm authority="marcrelator" type="text">editor</roleTerm>
</role>
</name>
<name type="personal">
<namePart type="given">Chen</namePart>
<namePart type="family">Zhao</namePart>
<role>
<roleTerm authority="marcrelator" type="text">editor</roleTerm>
</role>
</name>
<name type="personal">
<namePart type="given">Isabelle</namePart>
<namePart type="family">Augenstein</namePart>
<role>
<roleTerm authority="marcrelator" type="text">editor</roleTerm>
</role>
</name>
<name type="personal">
<namePart type="given">Anna</namePart>
<namePart type="family">Rogers</namePart>
<role>
<roleTerm authority="marcrelator" type="text">editor</roleTerm>
</role>
</name>
<name type="personal">
<namePart type="given">Kyunghyun</namePart>
<namePart type="family">Cho</namePart>
<role>
<roleTerm authority="marcrelator" type="text">editor</roleTerm>
</role>
</name>
<name type="personal">
<namePart type="given">Edward</namePart>
<namePart type="family">Grefenstette</namePart>
<role>
<roleTerm authority="marcrelator" type="text">editor</roleTerm>
</role>
</name>
<name type="personal">
<namePart type="given">Lena</namePart>
<namePart type="family">Voita</namePart>
<role>
<roleTerm authority="marcrelator" type="text">editor</roleTerm>
</role>
</name>
<originInfo>
<publisher>Association for Computational Linguistics</publisher>
<place>
<placeTerm type="text">Toronto, Canada</placeTerm>
</place>
</originInfo>
<genre authority="marcgt">conference publication</genre>
</relatedItem>
<abstract>We propose a novel task-agnostic in-domain pre-training method that sits between generic pre-training and fine-tuning. Our approach selectively masks in-domain keywords, i.e., words that provide a compact representation of the target domain. We identify such keywords using KeyBERT (Grootendorst, 2020). We evaluate our approach using six different settings: three datasets combined with two distinct pre-trained language models (PLMs). Our results reveal that the fine-tuned PLMs adapted using our in-domain pre-training strategy outperform PLMs that used in-domain pre-training with random masking as well as those that followed the common pre-train-then-fine-tune paradigm. Further, the overhead of identifying in-domain keywords is reasonable, e.g., 7-15% of the pre-training time (for two epochs) for BERT Large (Devlin et al., 2019).</abstract>
<identifier type="citekey">golchin-etal-2023-mask</identifier>
<identifier type="doi">10.18653/v1/2023.repl4nlp-1.2</identifier>
<location>
<url>https://aclanthology.org/2023.repl4nlp-1.2</url>
</location>
<part>
<date>2023-07</date>
<extent unit="page">
<start>13</start>
<end>21</end>
</extent>
</part>
</mods>
</modsCollection>
%0 Conference Proceedings
%T Do not Mask Randomly: Effective Domain-adaptive Pre-training by Masking In-domain Keywords
%A Golchin, Shahriar
%A Surdeanu, Mihai
%A Tavabi, Nazgol
%A Kiapour, Ata
%Y Can, Burcu
%Y Mozes, Maximilian
%Y Cahyawijaya, Samuel
%Y Saphra, Naomi
%Y Kassner, Nora
%Y Ravfogel, Shauli
%Y Ravichander, Abhilasha
%Y Zhao, Chen
%Y Augenstein, Isabelle
%Y Rogers, Anna
%Y Cho, Kyunghyun
%Y Grefenstette, Edward
%Y Voita, Lena
%S Proceedings of the 8th Workshop on Representation Learning for NLP (RepL4NLP 2023)
%D 2023
%8 July
%I Association for Computational Linguistics
%C Toronto, Canada
%F golchin-etal-2023-mask
%X We propose a novel task-agnostic in-domain pre-training method that sits between generic pre-training and fine-tuning. Our approach selectively masks in-domain keywords, i.e., words that provide a compact representation of the target domain. We identify such keywords using KeyBERT (Grootendorst, 2020). We evaluate our approach using six different settings: three datasets combined with two distinct pre-trained language models (PLMs). Our results reveal that the fine-tuned PLMs adapted using our in-domain pre-training strategy outperform PLMs that used in-domain pre-training with random masking as well as those that followed the common pre-train-then-fine-tune paradigm. Further, the overhead of identifying in-domain keywords is reasonable, e.g., 7-15% of the pre-training time (for two epochs) for BERT Large (Devlin et al., 2019).
%R 10.18653/v1/2023.repl4nlp-1.2
%U https://aclanthology.org/2023.repl4nlp-1.2
%U https://doi.org/10.18653/v1/2023.repl4nlp-1.2
%P 13-21
Markdown (Informal)
[Do not Mask Randomly: Effective Domain-adaptive Pre-training by Masking In-domain Keywords](https://aclanthology.org/2023.repl4nlp-1.2) (Golchin et al., RepL4NLP 2023)
ACL