Enhancing multiple-choice question answering through sequential fine-tuning and Curriculum Learning strategies

dc.authoridYiğit, Gülsüm/0000-0001-7010-169X
dc.authorwosidYiğit, Gülsüm/IVU-8380-2023
dc.contributor.authorYiğit, Gülsüm
dc.contributor.authorAmasyali, Mehmet Fatih
dc.date.accessioned2023-10-19T15:12:48Z
dc.date.available2023-10-19T15:12:48Z
dc.date.issued2023
dc.department-temp[Yigit, Gulsum] Kadir Has Univ, Dept Comp Engn, Istanbul, Turkiye; [Yigit, Gulsum; Amasyali, Mehmet Fatih] Yildiz Tech Univ, Dept Comp Engn, Istanbul, Turkiyeen_US
dc.description.abstractWith the transformer-based pre-trained language models, multiple-choice question answering (MCQA) systems can reach a particular level of performance. This study focuses on inheriting the benefits of contextualized language representations acquired by language models and transferring and sharing information among MCQA datasets. In this work, a method called multi-stage-fine-tuning considering the Curriculum Learning strategy is presented, which proposes sequencing not training samples, but the source datasets in a meaningful order, not randomized. Consequently, an extensive series of experiments over various MCQA datasets shows that the proposed method reaches remarkable performance enhancements than classical fine-tuning over picked baselines T5 and RoBERTa. Moreover, the experiments are conducted on merged source datasets, and the proposed method achieves improved performance. This study shows that increasing the number of source datasets and even using some small-scale datasets helps build well-generalized models. Moreover, having a higher similarity between source datasets and target also plays a vital role in the performance.en_US
dc.description.sponsorshipScientific and Technological Research Council of Turkey (TUBITAK) [120E100]; TUBITAK national fellowship program for PhD studies [BIDEB 2211/A]en_US
dc.description.sponsorshipThis research is supported by The Scientific and Technological Research Council of Turkey (TUBITAK) in part of the project with 120E100 Grant Number. G. Yigit is supported by TUBITAK - BIDEB 2211/A national fellowship program for PhD studies.en_US
dc.identifier.citation0
dc.identifier.doi10.1007/s10115-023-01918-2en_US
dc.identifier.endpage5042en_US
dc.identifier.issn0219-1377
dc.identifier.issn0219-3116
dc.identifier.issue11en_US
dc.identifier.scopus2-s2.0-85164111490en_US
dc.identifier.scopusqualityQ1
dc.identifier.startpage5025en_US
dc.identifier.urihttps://doi.org/10.1007/s10115-023-01918-2
dc.identifier.urihttps://hdl.handle.net/20.500.12469/5536
dc.identifier.volume65en_US
dc.identifier.wosWOS:001023370300001en_US
dc.identifier.wosqualityQ2
dc.khas20231019-WoSen_US
dc.language.isoenen_US
dc.publisherSpringer London Ltden_US
dc.relation.ispartofKnowledge and Information Systemsen_US
dc.relation.publicationcategoryMakale - Uluslararası Hakemli Dergi - Kurum Öğretim Elemanıen_US
dc.rightsinfo:eu-repo/semantics/closedAccessen_US
dc.subjectMCQAen_US
dc.subjectT5en_US
dc.subjectCommonsenseEn_Us
dc.subjectRoBERTaen_US
dc.subjectFine-tuningen_US
dc.subjectCommonsense
dc.subjectCurriculum-learningen_US
dc.titleEnhancing multiple-choice question answering through sequential fine-tuning and Curriculum Learning strategiesen_US
dc.typeArticleen_US
dspace.entity.typePublication
relation.isAuthorOfPublication363c092e-cd4b-400e-8261-ca5b99b1bea9
relation.isAuthorOfPublication.latestForDiscovery363c092e-cd4b-400e-8261-ca5b99b1bea9

Files

Original bundle

Now showing 1 - 1 of 1
No Thumbnail Available
Name:
5536.pdf
Size:
911.46 KB
Format:
Adobe Portable Document Format
Description:
Tam Metin / Full Text