C2PO: Diagnosing and Disentangling Bias Shortcuts in LLMs

Bias in Large Language Models (LLMs) poses significant risks to trustworthiness, manifesting primarily as stereotypical biases (e.g., gender or racial stereotypes) and structural biases (e.g., lexical overlap or position preferences). However, prior paradigms typically address these in isolation, often mitigating one at the expense of exacerbating the other. To address this, we conduct a systematic exploration of these reasoning failures and identify a primary inducement: the latent spurious feature correlations within the input that drive these erroneous reasoning shortcuts. Driven by these findings, we introduce Causal-Contrastive Preference Optimization (C2PO), a unified alignment framework designed to tackle these specific failures by simultaneously discovering and suppressing these correlations directly within the optimization process. Specifically, C2PO leverages causal counterfactual signals to isolate bias-inducing features from valid reasoning paths, and employs a fairness-sensitive preference update mechanism to dynamically evaluate logit-level contributions and suppress shortcut features. Extensive experiments across multiple benchmarks covering stereotypical bias (BBQ, Unqover), structural bias (MNLI, HANS, Chatbot, MT-Bench), out-of-domain fairness (StereoSet, WinoBias), and general utility (MMLU, GSM8K) demonstrate that C2PO effectively mitigates stereotypical and structural biases while preserving robust general reasoning capabilities.

FairGSE: Fairness-Aware Graph Neural Network without High False Positive Rates

Graph neural networks (GNNs) have emerged as the mainstream paradigm for graph representation learning due to their effective message aggregation. However, this advantage also amplifies biases inherent in graph topology, raising fairness concerns. Existing fairness-aware GNNs provide satisfactory performance on fairness metrics such as Statistical Parity and Equal Opportunity while maintaining acceptable accuracy trade-offs. Unfortunately, we observe that this pursuit of fairness metrics neglects the GNN's ability to predict negative labels, which renders their predictions with extremely high False Positive Rates (FPR), resulting in negative effects in high-risk scenarios. To this end, we advocate that classification performance should be carefully calibrated while improving fairness, rather than simply constraining accuracy loss. Furthermore, we propose Fair GNN via Structural Entropy (\textbf{FairGSE}), a novel framework that maximizes two-dimensional structural entropy (2D-SE) to improve fairness without neglecting false positives. Experiments on several real-world datasets show FairGSE reduces FPR by 39\% vs. state-of-the-art fairness-aware GNNs, with comparable fairness improvement.