药食同源类拼多多直播带货违法监测预警数据
收藏资源简介：
对采集的拼多多平台带货品类为药食同源类的达人直播视频内容进行转译分析，对达人口播语言内容进行处理、分析，根据达人在直播过程中对预先设置的违规敏感词（比如：彻底根治、绝对安全、无任何副作用、快速见效、立竿见影、包治百病、100%纯天然、预防疾病、抗癌、抗衰老、政府推荐等）违反的次数和频率，依据触发条件规则提出警告或处理。为服务辖区市场监督局管理区域内规范企业拼多多直播行为，提供数据支持。将采集完成的直播视频进行进行预处理，第一步：基于原始视频文件，以最大10分钟单位对原始视频进行切片。第二步：对于已完成的切片视频，进行视频内容转语音操作。第三步：对于已完成视频转语音操作的切片，进行语音转文本操作。第四步：使用OCR技术对原始视频中抓取的图片进行文字提取操作。第五步：将所得到的文字内容与违法预警关键词库进行匹配。最终运用多标准决策分析模型，对主播在直播过程中出现的违规语句进行分析计算，得出违法预警值和是否预警判断。 违法预警值=（违法预警单关键词命中次数*0.25）+（违法预警组合关键词命中次数* 0.3）+（图片识别命中预警组合关键词个数*0.35）+（直播间近一个月历史违规记录数*0.1） 通过公式计算出最终违法预警值，违法预警值 ≤1 时，不触发预警提示，违法预警值 ＞1 时触发违法预警提示。
This dataset is constructed via the transcription and analysis of influencer live streaming videos for medicinal and edible product categories collected from the Pinduoduo e-commerce platform. We process and analyze the spoken language content of influencers during live broadcasts, and issue warnings or take regulatory actions in accordance with pre-defined triggering rules based on the frequency and count of violations of pre-set sensitive prohibited keywords. The prohibited keywords include but are not limited to: "completely cure", "absolutely safe", "no side effects whatsoever", "rapidly effective", "instantly effective", "cure all diseases", "100% pure natural", "prevent diseases", "anti-cancer", "anti-aging", "government recommended", etc. This work provides data support for the Market Supervision Administration under its jurisdiction to regulate the live streaming behaviors of enterprises on the Pinduoduo platform. The collected live streaming videos undergo the following preprocessing procedures: 1. Slice the original video files into segments with a maximum duration of 10 minutes; 2. Convert the content of each sliced video into audio tracks; 3. Perform speech-to-text (STT) conversion on the extracted audio tracks from the sliced videos; 4. Extract text from images captured in the original videos using Optical Character Recognition (OCR) technology; 5. Match all the obtained text content against the pre-built prohibited early-warning keyword database. Finally, a multi-criteria decision analysis (MCDA) model is utilized to analyze and calculate the violation statements made by the live stream host, so as to derive the violation early warning score and determine whether to trigger an early warning. The violation early warning score is calculated via the following formula: Violation Early Warning Score = (Number of single prohibited warning keyword hits * 0.25) + (Number of combined prohibited warning keyword hits * 0.3) + (Number of combined prohibited warning keywords detected via image recognition * 0.35) + (Number of the host's historical violation records in the past month * 0.1) The early warning triggering rules are as follows: no early warning prompt will be triggered if the Violation Early Warning Score ≤ 1; a violation early warning prompt will be triggered if the Violation Early Warning Score > 1.
