SubmissionNumber#=%=#20
FinalPaperTitle#=%=#LinguIUTics at PsyDefDetect: Iterative Imbalance-Aware Fine-tuning of Qwen3-8B for Psychological Defense Mechanism Classification
ShortPaperTitle#=%=#
NumberOfPages#=%=#
CopyrightSigned#=%=#
JobTitle#==#
Organization#==#
Abstract#==#Detecting psychological defense mechanisms in conversational text remains a challenging clinical NLP problem. For the PsyDefDetect 2026 shared task (9-class utterance classification evaluated via macro F1), our team LinguIUTics ranks 3rd overall, achieving an F1-score of 0.4427 (2nd highest recall) in the all-class leaderboard and 4th overall in the positive-class leaderboard, achieving an F1-score of 0.3917, a 24.6% gain over the reported SOTA. BERT-family encoders and zero-shot LLMs proved ineffective on rare classes due to severe class imbalance, leading us to QLoRA fine-tuning of Qwen3-8B. We leverage three key strategies: grouped stratified cross-validation (preventing leakage), minority-class lexical augmentation, and a post-processing pipeline with logit bias tuning and ensemble blending. These yield a +3.69 F1 improvement, driving the critical "Unclear" class to F1 = 0.797.
Author{1}{Firstname}#=%=#Shefayat E. Shams
Author{1}{Lastname}#=%=#Adib
Author{1}{Orcid}#=%=#
Author{1}{Email}#=%=#shefayatadib@iut-dhaka.edu
Author{1}{Affiliation}#=%=#Islamic University of Technology
Author{2}{Firstname}#=%=#Ahmed Alfey
Author{2}{Lastname}#=%=#Sani
Author{2}{Orcid}#=%=#
Author{2}{Email}#=%=#ahmedalfey@iut-dhaka.edu
Author{2}{Affiliation}#=%=#Islamic University of Technology
Author{3}{Firstname}#=%=#Md Hasibur Rahman
Author{3}{Lastname}#=%=#Alif
Author{3}{Orcid}#=%=#
Author{3}{Email}#=%=#hasiburrahman21@iut-dhaka.edu
Author{3}{Affiliation}#=%=#Islamic University of Technology
Author{4}{Firstname}#=%=#Ajwad
Author{4}{Lastname}#=%=#Abrar
Author{4}{Username}#=%=#ajwad-abrar
Author{4}{Orcid}#=%=#
Author{4}{Email}#=%=#ajwadabrar@iut-dhaka.edu
Author{4}{Affiliation}#=%=#IUT

==========
èéáğö