The POLAR @ SemEval-2026 Task 9 deals with the detection of online polarization in a variety of multilingual and multicultural environments. Our team participated in Subtask 1 of the POLAR @ SemEval-2026 Task 9, which mainly deals with binary classification of textual sequences for the detection of polarized stances. In this paper, we proposed a strong classification system for Bengali language based on fine-tuning the BanglaBERT Large model. The methodology used here involves a stratified five-fold cross-validation approach along with a performance-weighted ensemble method, combined with temperature scaling probability calibration and a set of lexical post-processing rules.

Psychological defense mechanisms play a cru-cial role in shaping human responses duringemotionally charged conversations, yet remainunderexplored in natural language processing.In this work, we address the PSYDEFCONVshared task, which involves classifying defensemechanisms in multi-turn dialogues using clin-ically grounded annotations based on the De-fense Mechanism Rating Scales (DMRS). Wepropose a generative supervised fine-tuningframework that reformulates the task as con-ditional text generation. A pre-trained causallanguage model (Gemma-2-2B) is adapted us-ing parameter-efficient fine-tuning (PEFT) with4-bit quantization, enabling efficient trainingunder limited computational resources. To han-dle class imbalance, we apply random oversam-pling, and we design a prompt-based input rep-resentation to incorporate conversational con-text effectively. Experimental results demon-strate that our generative approach is compet-itive with discriminative baselines while of-fering improved flexibility in modeling sub-tle and context-dependent defensive behaviors.The findings highlight the potential of genera-tive large language models for psychologicallygrounded dialogue understanding tasks.