WIPI/deceptive_patterns_synthetic

该数据集包含了从屏幕截图（如文本、类型、坐标、颜色等）中提取的UI元素特征及其对应的标签，这些标签将它们分类为non-deceptive或几种欺骗模式的类别（例如forced-action、interface-interference、sneaking）。此数据集的主要目的是作为知识蒸馏的大规模资源，捕捉了最先进的教师模型（Gemini）对给定UI元素为何具有欺骗性的分类和推理。This dataset consists of UI element features extracted from screenshots (text, type, coordinates, colors, etc.) and corresponding labels that classify them as either non-deceptive or one of several categories of deceptive patterns (e.g., forced-action, interface-interference, sneaking). The primary purpose of this dataset is to serve as a large-scale resource for knowledge distillation, capturing the classifications and reasoning of a state-of-the-art teacher model (Gemini) for why a given UI element is deceptive or not.