Abstract This research paper proposed a multilingual multi-label classification of online polarization types via weighted transformer fine-tuning and adaptive per-label threshold optimization (MMCOPT). Our task is to classify social media posts according to a given set of five labels. A post could be deemed to be politically, racially, religiously, or gender/sexually polarizing, or fall into the category of other. We incorporate a distilbert-base-multilingualcased model and attach a two-layer MLP head. We also use a class-imbalance-weighted binary cross-entropy loss and optimize thresholds for each class to improve the validation micro-F1 score. Our training set is drawn from the POLAR benchmark, the first large multilingual polarization dataset that includes posts from seven languages and multiple social media platforms. MMCOPT’s best internal validation micro-F1 score is 0.7855, and its macro-F1 score is 0.7749. Our model (team username: asbichi362) is ranked on the official Codabench leaderboard and shows competitive results across 22 language tracks of the research project multilingual polarization type classification, with its best results in Hindi (0.7429) and Urdu (0.7073).

This paper presents our findings of the Multilingual Shared Task on Hallucinations and Related Observable Overgeneration Mistakes, MU-SHROOM, which focuses on identifying hallucinations and related overgeneration errors in large language models (LLMs). The shared task involves detecting specific text spans that constitute hallucinations in the outputs generated by LLMs in 14 languages. To address this task, we aim to provide a nuanced, model-aware understanding of hallucination occurrences and severity in English. We used natural language inference and fine-tuned a ModernBERT model using a synthetic dataset of 400 samples, achieving an Intersection over Union (IoU) score of 0.032 and a correlation score of 0.422. These results indicate a moderately positive correlation between the model’s confidence scores and the actual presence of hallucinations. The IoU score indicates that our modelhas a relatively low overlap between the predicted hallucination span and the truth annotation. The performance is unsurprising, given the intricate nature of hallucination detection. Hallucinations often manifest subtly, relying on context, making pinpointing their exact boundaries formidable.