软餐获悉，荷兰媒体 NOS（Nederlandse Omroep Stichting，荷兰国家广播基金会）开发了两种机器学习模型来检测聊天应用 Telegram 上的毒品广告。已发现了数百万条荷兰供应商提供毒品的信息。这涉及 9900 个不同的供应商，据说有些账户已经活跃多年。在 NOS 的测试数据集中，模型分别有 97% 和 96% 的时间是正确的。该媒体已将这两种模型合并并发布在 Huggingface 上。NOS 称，检测结果给人的印象是 Telegram 对这些信息 “几乎没有干预”。确实，Telegram 通常不会遵守提供有关可疑用户信息的请求。该公司也不会主动追踪这些组织。而 Telegram 发言人向 NOS 声称，该公司确实对有害信息进行了主动监控，并对用户的投诉做出了回应。
The Dutch media organization NOS (Nederlandse Omroep Stichting, Dutch Broadcasting Foundation) has developed two machine learning models to detect drug advertisements on the chat application Telegram. Millions of messages from Dutch suppliers offering drugs have been discovered. Some accounts are reported to have been active for several years, involving 9,900 different suppliers. In NOS’s test dataset, the models were found to be correct 97% and 96% of the time, respectively. These two models have been merged by NOS and released on Huggingface. NOS stated that the impression from the detection results is that Telegram has “hardly any intervention” regarding this information. Indeed, Telegram typically does not comply with requests to provide information about suspicious users, nor does it actively track these organizations. However, a Telegram spokesperson told NOS that the company does actively monitor harmful information and responds to user complaints.
微信扫一扫