@inproceedings{taechoyotin-acuna-2024-misti,
title = "{MISTI}: Metadata-Informed Scientific Text and Image Representation through Contrastive Learning",
author = "Taechoyotin, Pawin and
Acuna, Daniel",
editor = "Ghosal, Tirthankar and
Singh, Amanpreet and
Waard, Anita and
Mayr, Philipp and
Naik, Aakanksha and
Weller, Orion and
Lee, Yoonjoo and
Shen, Shannon and
Qin, Yanxia",
booktitle = "Proceedings of the Fourth Workshop on Scholarly Document Processing (SDP 2024)",
month = aug,
year = "2024",
address = "Bangkok, Thailand",
publisher = "Association for Computational Linguistics",
url = "https://aclanthology.org/2024.sdp-1.15",
pages = "155--164",
abstract = "In scientific publications, automatic representations of figures and their captions can be used in NLP, computer vision, and information retrieval tasks. Contrastive learning has proven effective for creating such joint representations for natural scenes, but its application to scientific imagery and descriptions remains under-explored. Recent open-access publication datasets provide an opportunity to understand the effectiveness of this technique as well as evaluate the usefulness of additional metadata, which are available only in the scientific context. Here, we introduce MISTI, a novel model that uses contrastive learning to simultaneously learn the representation of figures, captions, and metadata, such as a paper{'}s title, sections, and curated concepts from the PubMed Open Access Subset. We evaluate our model on multiple information retrieval tasks, showing substantial improvements over baseline models. Notably, incorporating metadata doubled retrieval performance, achieving a Recall@1 of 30{\%} on a 70K-item caption retrieval task. We qualitatively explore how metadata can be used to strategically retrieve distinctive representations of the same concept but for different sections, such as introduction and results. Additionally, we show that our model seamlessly handles out-of-domain tasks related to image segmentation. We share our dataset and methods (https://github.com/Khempawin/scientific-image-caption-pair/tree/section-attr) and outline future research directions.",
}
<?xml version="1.0" encoding="UTF-8"?>
<modsCollection xmlns="http://www.loc.gov/mods/v3">
<mods ID="taechoyotin-acuna-2024-misti">
<titleInfo>
<title>MISTI: Metadata-Informed Scientific Text and Image Representation through Contrastive Learning</title>
</titleInfo>
<name type="personal">
<namePart type="given">Pawin</namePart>
<namePart type="family">Taechoyotin</namePart>
<role>
<roleTerm authority="marcrelator" type="text">author</roleTerm>
</role>
</name>
<name type="personal">
<namePart type="given">Daniel</namePart>
<namePart type="family">Acuna</namePart>
<role>
<roleTerm authority="marcrelator" type="text">author</roleTerm>
</role>
</name>
<originInfo>
<dateIssued>2024-08</dateIssued>
</originInfo>
<typeOfResource>text</typeOfResource>
<relatedItem type="host">
<titleInfo>
<title>Proceedings of the Fourth Workshop on Scholarly Document Processing (SDP 2024)</title>
</titleInfo>
<name type="personal">
<namePart type="given">Tirthankar</namePart>
<namePart type="family">Ghosal</namePart>
<role>
<roleTerm authority="marcrelator" type="text">editor</roleTerm>
</role>
</name>
<name type="personal">
<namePart type="given">Amanpreet</namePart>
<namePart type="family">Singh</namePart>
<role>
<roleTerm authority="marcrelator" type="text">editor</roleTerm>
</role>
</name>
<name type="personal">
<namePart type="given">Anita</namePart>
<namePart type="family">Waard</namePart>
<role>
<roleTerm authority="marcrelator" type="text">editor</roleTerm>
</role>
</name>
<name type="personal">
<namePart type="given">Philipp</namePart>
<namePart type="family">Mayr</namePart>
<role>
<roleTerm authority="marcrelator" type="text">editor</roleTerm>
</role>
</name>
<name type="personal">
<namePart type="given">Aakanksha</namePart>
<namePart type="family">Naik</namePart>
<role>
<roleTerm authority="marcrelator" type="text">editor</roleTerm>
</role>
</name>
<name type="personal">
<namePart type="given">Orion</namePart>
<namePart type="family">Weller</namePart>
<role>
<roleTerm authority="marcrelator" type="text">editor</roleTerm>
</role>
</name>
<name type="personal">
<namePart type="given">Yoonjoo</namePart>
<namePart type="family">Lee</namePart>
<role>
<roleTerm authority="marcrelator" type="text">editor</roleTerm>
</role>
</name>
<name type="personal">
<namePart type="given">Shannon</namePart>
<namePart type="family">Shen</namePart>
<role>
<roleTerm authority="marcrelator" type="text">editor</roleTerm>
</role>
</name>
<name type="personal">
<namePart type="given">Yanxia</namePart>
<namePart type="family">Qin</namePart>
<role>
<roleTerm authority="marcrelator" type="text">editor</roleTerm>
</role>
</name>
<originInfo>
<publisher>Association for Computational Linguistics</publisher>
<place>
<placeTerm type="text">Bangkok, Thailand</placeTerm>
</place>
</originInfo>
<genre authority="marcgt">conference publication</genre>
</relatedItem>
<abstract>In scientific publications, automatic representations of figures and their captions can be used in NLP, computer vision, and information retrieval tasks. Contrastive learning has proven effective for creating such joint representations for natural scenes, but its application to scientific imagery and descriptions remains under-explored. Recent open-access publication datasets provide an opportunity to understand the effectiveness of this technique as well as evaluate the usefulness of additional metadata, which are available only in the scientific context. Here, we introduce MISTI, a novel model that uses contrastive learning to simultaneously learn the representation of figures, captions, and metadata, such as a paper’s title, sections, and curated concepts from the PubMed Open Access Subset. We evaluate our model on multiple information retrieval tasks, showing substantial improvements over baseline models. Notably, incorporating metadata doubled retrieval performance, achieving a Recall@1 of 30% on a 70K-item caption retrieval task. We qualitatively explore how metadata can be used to strategically retrieve distinctive representations of the same concept but for different sections, such as introduction and results. Additionally, we show that our model seamlessly handles out-of-domain tasks related to image segmentation. We share our dataset and methods (https://github.com/Khempawin/scientific-image-caption-pair/tree/section-attr) and outline future research directions.</abstract>
<identifier type="citekey">taechoyotin-acuna-2024-misti</identifier>
<location>
<url>https://aclanthology.org/2024.sdp-1.15</url>
</location>
<part>
<date>2024-08</date>
<extent unit="page">
<start>155</start>
<end>164</end>
</extent>
</part>
</mods>
</modsCollection>
%0 Conference Proceedings
%T MISTI: Metadata-Informed Scientific Text and Image Representation through Contrastive Learning
%A Taechoyotin, Pawin
%A Acuna, Daniel
%Y Ghosal, Tirthankar
%Y Singh, Amanpreet
%Y Waard, Anita
%Y Mayr, Philipp
%Y Naik, Aakanksha
%Y Weller, Orion
%Y Lee, Yoonjoo
%Y Shen, Shannon
%Y Qin, Yanxia
%S Proceedings of the Fourth Workshop on Scholarly Document Processing (SDP 2024)
%D 2024
%8 August
%I Association for Computational Linguistics
%C Bangkok, Thailand
%F taechoyotin-acuna-2024-misti
%X In scientific publications, automatic representations of figures and their captions can be used in NLP, computer vision, and information retrieval tasks. Contrastive learning has proven effective for creating such joint representations for natural scenes, but its application to scientific imagery and descriptions remains under-explored. Recent open-access publication datasets provide an opportunity to understand the effectiveness of this technique as well as evaluate the usefulness of additional metadata, which are available only in the scientific context. Here, we introduce MISTI, a novel model that uses contrastive learning to simultaneously learn the representation of figures, captions, and metadata, such as a paper’s title, sections, and curated concepts from the PubMed Open Access Subset. We evaluate our model on multiple information retrieval tasks, showing substantial improvements over baseline models. Notably, incorporating metadata doubled retrieval performance, achieving a Recall@1 of 30% on a 70K-item caption retrieval task. We qualitatively explore how metadata can be used to strategically retrieve distinctive representations of the same concept but for different sections, such as introduction and results. Additionally, we show that our model seamlessly handles out-of-domain tasks related to image segmentation. We share our dataset and methods (https://github.com/Khempawin/scientific-image-caption-pair/tree/section-attr) and outline future research directions.
%U https://aclanthology.org/2024.sdp-1.15
%P 155-164
Markdown (Informal)
[MISTI: Metadata-Informed Scientific Text and Image Representation through Contrastive Learning](https://aclanthology.org/2024.sdp-1.15) (Taechoyotin & Acuna, sdp-WS 2024)
ACL