CSECU-DSG at SemEval-2026 Task 6: Imbalance-Aware Transformers for Unmasking Political Question Evasions
Year: 2026
Type: conference-paper
Abstract: Clarity-Level Classification predicts the degree of clarity of a response to a query.It is essential to the advancement of many NLP applications, including conversational AI, customer support automation, and instructional technology.However, it is challenging to assess the clarity of an answer due to unclear wording, incomplete answers, and the contextual dependency in Q/A.In this paper, we present our approach to the shared task on clarity classification introduced in SemEval 2026 Task 6.We formulate the problem as both a regression and multi-class classification task over question-answer pairs and propose a transformer-based architecture to model their contextual interactions.Our system leverages a fine-tuned DeBERTa-v3-base backbone to capture nuanced semantic relationships between questions and answers.To mitigate class imbalance, we incorporate class-weighted loss functions and apply data-level oversampling techniques.Experimental results demonstrate that our approach achieves competitive performance in the shared task, highlighting the effectiveness of transformer-based contextual modeling combined with imbalance-aware training strategies.
Institution: University of Chittagong
Language: English
Topic: Topic Modeling
Subfield: Artificial Intelligence
Field: Computer Science
Domain: Physical Sciences
Open Access status: hybrid
OpenAlex ID: W7166878251