InsBERT: Word importance from artificial insertions
The result's identifiers
Result code in IS VaVaI
<a href="https://www.isvavai.cz/riv?ss=detail&h=RIV%2F00216208%3A11320%2F24%3A10492892" target="_blank" >RIV/00216208:11320/24:10492892 - isvavai.cz</a>
Result on the web
<a href="https://ceur-ws.org/Vol-3792/paper11.pdf" target="_blank" >https://ceur-ws.org/Vol-3792/paper11.pdf</a>
DOI - Digital Object Identifier
—
Alternative languages
Result language
angličtina
Original language name
InsBERT: Word importance from artificial insertions
Original language description
We investigate the quantification of word importance by introducing a novel self-supervised task that modifies masked language modeling. Instead of predicting masked words, our approach involves learning to identify which words were inserted. We hypothesize that resulting models will predict a higher likelihood of insertion for less important words. We experiment with two different insertion strategies: the List Inserting Method (LIM) and the BERT Inserting Method (BIM). We outline the process for gathering manually estimated word importance data and describe the construction of a dataset for evaluating our methods. Our results indicate that our modified language modeling surpasses baselines and is competitive with existing research in assessing word importance.
Czech name
—
Czech description
—
Classification
Type
D - Article in proceedings
CEP classification
—
OECD FORD branch
10201 - Computer sciences, information science, bioinformathics (hardware development to be 2.2, social aspect to be 5.8)
Result continuities
Project
<a href="/en/project/GX19-26934X" target="_blank" >GX19-26934X: Neural Representations in Multi-modal and Multi-lingual Modeling</a><br>
Continuities
P - Projekt vyzkumu a vyvoje financovany z verejnych zdroju (s odkazem do CEP)
Others
Publication year
2024
Confidentiality
S - Úplné a pravdivé údaje o projektu nepodléhají ochraně podle zvláštních právních předpisů
Data specific for result type
Article name in the collection
Proceedings of the 24th Conference Information Technologies – Applications and Theory (ITAT 2024)
ISBN
—
ISSN
1613-0073
e-ISSN
—
Number of pages
11
Pages from-to
96-106
Publisher name
CEUR-WS.org
Place of publication
Košice, Slovakia
Event location
Drienica, Slovakia
Event date
Sep 20, 2024
Type of event by nationality
CST - Celostátní akce
UT code for WoS article
—