Counterfactual debiasing is often used to reduce reliance on identity-linked shortcuts in harmful language understanding. We show that the same intervention can be harmful when identity mentions are not shortcuts but label-defining evidence. In multi-task cyberbullying modeling, identity cues may be spurious for cyberbullying detection (CD) and sentiment analysis (SA), yet necessary for fine-grained target identification (TI). We propose task-aware routing, which sends CD and SA through a debiased representation while preserving the original representation for TI. We evaluate this approach on the English-translated BullyExplain benchmark, HateXplain, and an external Measuring Hate Speech diagnostic. Experiments show that task-aware routing improves TI, while auxiliary detection and sentiment performance remain largely unchanged where applicable. Ablation studies attribute this improvement to representation routing that preserves identity-bearing evidence for TI rather than to class-weighting choices, with reduced-architecture diagnostics suggesting that the effect is not specific to the graph-enhanced model. These results support a task-conditional view of counterfactual debiasing: the same identity cue can warrant suppression in one task and preservation in another.

Fact verification is an essential tool to mitigate the spread of false information online, which has gained a widespread attention recently. However, a fact verification in the question-answering dialogue is still underexplored. In this paper, we propose a neural network based approach called question-answering dialogue based fact verification with mixture of experts (QaDialMoE). It exploits questions and evidence effectively in the verification process and can significantly improve the performance of fact verification. Specifically, we exploit the mixture of experts to focus on various interactions among responses, questions and evidence. A manager with an attention guidance module is implemented to guide the training of experts and assign a reasonable attention score to each expert. A prompt module is developed to generate synthetic questions that make our approach more generalizable. Finally, we evaluate the QaDialMoE and conduct a comparative study on three benchmark datasets. The experimental results demonstrate that our QaDialMoE outperforms previous approaches by a large margin and achieves new state-of-the-art results on all benchmarks. This includes the accuracy improvements on the HEALTHVER as 84.26%, the FAVIQ A dev set as 78.7%, the FAVIQ R dev set as 86.1%, test set as 86.0%, and the COLLOQUIAL as 89.5%. To our best knowledge, this is the first work to investigate a question-answering dialogue based fact verification, and achieves new state-of-the-art results on various benchmark datasets.