SVP Content Moderation Benchmark

该数据集是一个针对短视频平台内容审核的基准数据集，它包含了真实用户和评审者的反馈，涵盖了15种违规视频类型。该数据集被用于训练包括所提出的KuaiMod方法在内的多种方法，重点关注二分类和多分类的性能表现。数据集规模为24,562个样本，其任务是内容审核。This dataset is a benchmark dataset for short-video platform content moderation. It includes feedback from real users and human annotators, covering 15 types of violating video content. This dataset has been utilized to train multiple methods including the proposed KuaiMod approach, with a focus on the performance of binary classification and multi-classification tasks. Comprising 24,562 samples in total, its designated task is content moderation.

搜集汇总

背景与挑战

背景概述

SVP Content Moderation Benchmark（又名KuaiMod）是首个针对SVP生态系统治理的基准数据集，完全基于快手平台的内容进行人工标注，专门用于内容审核任务的评估和优化。