Learning Sexism Detection Using Multi-Agent Perspectivist Preference Optimization

When people label text for sexism, they often disagree, and not because some of them are wrong: they genuinely perceive sexism differently. Most NLP systems discard this disagreement by collapsing it into a majority vote. We propose the Multi-Agent Perspectivist Preference Optimization (MAP-PO) framework to keep these different perspectives. On the EXIST 2024 dataset of labeled English and Spanish tweets, we first cluster annotators by their labeling behavior rather than their demographic attributes. We then fine-tune one Large Language Model agent per cluster to reproduce that cluster's annotation behavior, and coordinate the agents with preference optimization that combines individual and team-level rewards. We evaluate MAP-PO in four settings defined by two languages and two backbone language models, asking whether each agent reproduces the annotations of its own cluster and whether the agents together reproduce the majority label. Two findings hold in all four settings. First, without fine-tuning the agents behave almost identically, so cluster-specific training is necessary. Second, we show that training each agent only on the labels of its own cluster pushes the agents far beyond the clusters they should represent, while adding a shared team-level training signal consistently keeps each agent calibrated to its cluster.

BioSentinel at EXIST 2026: Soft-Label Optimization with XLM-RoBERTa for Sexism Intent Classification in Memes

This paper describes the BioSentinel team's participation in EXIST 2026 Task 2.2: Source Intention in Memes, part of the CLEF 2026 evaluation campaign. The task requires classifying the communicative intent behind memes as direct, judgemental, or no (non-sexist), under a Learning with Disagreement (Le-Wi-Di) paradigm that mandates both hard-label and soft-label (probability distribution) predictions. We present a text-centric approach built on xlm-roberta-base (270M parameters) trained with a composite loss function combining KL divergence on soft annotator distributions and weighted cross-entropy on hard labels. On the official test set, the system achieved an ICM-Soft-Norm of 0.3229 and ICM-Norm of 0.3778, with a hard F1-score of 0.4236, ranking 40th (out of 118 submissions) in the soft-soft evaluation and 49th (out of 187 submissions) in the hard-hard evaluation. We provide an analysis of the dataset characteristics, exploratory larger-architecture runs, and the role of annotator disagreement in shaping model design for subjective NLP tasks. Ablation results show that KL loss improves soft-label metrics, while CE loss improves hard-label accuracy. We also report a separate validation-set temperature analysis.

Reliability-Aware Sexism Detection: Combining DPO with Annotator Agreement and Token-Level Confidence Scoring

The detection of online sexism remains an open problem. Sexism detection is inherently subjective, yet most existing systems reduce multi-annotator labels to a single majority decision and treat all instances uniformly. This ignores two informative signals: annotator agreement and model uncertainty. We propose RA-DPO (Reliability-Aware Direct Preference Optimization), which integrates annotator agreement, model confidence, and a token-level uncertainty signal into a single reliability score. RA-DPO uses this score to select high-value preference pairs during training and to support inference-time abstention, which allows the model to trade coverage for accuracy. We evaluate RA-DPO on 6,920 multilingual posts from EXIST 2023, fine-tune OpenAI gpt-4o base via DPO, and validate on two open-weight 3B models (Llama, Qwen). Results show that training on the top 30% most reliable pairs matches full-data DPO, which indicates that reliability-aware selection can reduce training cost without sacrificing performance. At inference, selective prediction reaches 96.2% accuracy at 50% coverage in the true-agreement setting and 88.7% in the deployable predicted-agreement setting, both exceeding the 85.3% no-agreement baseline. These results suggest that accounting for annotation uncertainty is beneficial for both efficient training and reliable deployment in subjective classification.