PANC

该数据集名为PANC，包含了来自ChatCoder2的正面（诱骗行为）完整聊天记录，来自PAN12的负面（正常）聊天片段，以及将正面完整聊天分割成多个部分的正面片段。此外，该数据集显示出明显的类别不平衡，负面聊天数量是正面聊天的100倍以上。每个对话轮次都采用了策略级别的标注。该数据集的任务是针对在线诱骗行为的早期检测。This dataset is named PANC. It contains full-length positive (deceptive behavior) chat logs sourced from ChatCoder2, negative (normal) chat snippets from PAN12, as well as positive segments obtained by splitting the full-length positive chat logs into multiple parts. Notably, this dataset exhibits a significant class imbalance, with the number of negative chats being over 100 times that of positive ones. Each dialogue turn is annotated at the strategy level. The downstream task of this dataset is early detection of online deceptive behavior.

搜集汇总

背景与挑战

背景概述

该数据集用于在线诱骗行为的早期检测，包含正面（诱骗）和负面（正常）聊天记录，每个对话轮次采用策略级别标注，但存在严重的类别不平衡，负面样本数量是正面的100倍以上。