SENTI_RATIONAL情感可解释性数据集

“可信AI”的概念在近几年提出，并且逐渐成为全球共识。可信AI，就是保证AI系统的可解释性、鲁棒性和公平性，确保算法可被人信任。在情感分析技术的科学研究和产业落地中，可信情感分析系统需要具备较强的鲁棒性和可解释能力，也就是要求模型针对各种不同表达，既能给出精准的预测结果，还能给出其判断依据，从而保证模型的预测结果更加置信。

作为业界首个中文情感可信数据集，DuTrust基于人工标注的扰动数据和情感证据，全部来源于真实用户评论数据，同时测试模型的鲁棒性和可解释性，全面评估模型的可信能力。

The concept of "Trustworthy AI" was proposed in recent years and has gradually achieved global consensus. Trustworthy AI refers to ensuring the interpretability, robustness, and fairness of AI systems, thereby making algorithms trustworthy to humans. In the scientific research and industrial deployment of sentiment analysis technologies, a trustworthy sentiment analysis system should possess strong robustness and interpretability: specifically, the model must not only deliver accurate prediction results for diverse expressions, but also provide its judgment basis, thus enhancing the credibility of the model's predictions.

As the first Chinese trustworthy sentiment dataset in the industry, DuTrust is built upon manually annotated perturbed data and sentiment evidence, all of which are sourced from real user review data. It simultaneously tests the robustness and interpretability of models, and comprehensively evaluates the trustworthiness of AI models.