DeSIQ
收藏资源简介:
DeSIQ是一个用于社会智能理解的无偏见、更具挑战性的多模态问答基准数据集。该数据集由莫纳什大学的信息技术学院创建,通过对Social-IQ数据集进行扰动处理,旨在减少原数据集中的偏见。DeSIQ包含约21306条数据,每条数据包含视频、问题和多个选择答案,用于评估机器学习模型在复杂社会交互中的理解和推理能力。数据集的创建过程涉及对原数据集的答案选项进行替换,以确保答案的多样性和挑战性。DeSIQ的应用领域主要集中在社会智能研究,旨在解决现有AI系统在理解人类表达、意图和交互方面的局限性。
DeSIQ is an unbiased, more challenging multimodal question answering benchmark dataset for social intelligence understanding. It was developed by the School of Information Technology at Monash University via perturbing the original Social-IQ dataset, with the goal of mitigating biases inherent in the original dataset. DeSIQ contains approximately 21,306 instances, each of which comprises a video, a natural language question and multiple-choice answer options, and is used to evaluate the capacities of machine learning models for understanding and reasoning in complex social interactions. The dataset construction process involves replacing the answer options from the original Social-IQ dataset to ensure the diversity and challenging nature of the answer choices. The main application areas of DeSIQ are focused on social intelligence research, aiming to address the limitations of existing AI systems in comprehending human expressions, intentions and interpersonal interactions.




