|
所在平台: Udemy |
课程主页: https://www.udemy.com/course/insultbot/
课程评论：没有评论
**课程名称：** Let's build an insultbot **课程概述：** 本课程旨在探讨如何利用人工智能进行在线内容审核，特别是识别和过滤侮辱性言论。课程将使用 TensorFlow.js (JavaScript 机器学习库) 和一个预训练模型，以便轻松构建一个能够检测并标记不当内容的“侮辱机器人”。 **核心内容：** * **内容审核的重要性：** 讨论在线内容审核的挑战，以及“Napalm girl”事件等案例说明其复杂性。 * **人工智能在内容审核中的应用：** 介绍使用 TensorFlow.js 和预训练模型来识别侮辱性内容，该模型能够返回内容类型和相应的置信度分数。 * **技术栈：** * **TensorFlow.js：** 用于在浏览器端构建机器学习应用的 JavaScript 库。 * **Angular：** 一个用于构建前端应用的免费框架，无需服务器。 * **侮辱检测：** 重点关注直接侮辱（人身攻击、起侮辱性绰号）的识别，并将其作为一种模式识别问题来处理，通过模型输出的概率分数来衡量侮辱的程度。 * **免费且易于使用的方案：** 相较于需要API调用的ChatGPT，本课程提供了一个免费、代码量少且不需要专业知识的替代方案。 **最终目标：** 通过本课程，学习者将能够构建一个简单但有效的“侮辱机器人”，以辅助在线内容审核，并理解人工智能在这一领域的应用原理。
"even the prime minister of Norway herself, reposted the photo to Facebook, only to have it quickly removed." Custodians of the Internet (Tarleton Gillespie)A very nice application of artificial intelligence is on content moderation. Our goal is spotting whether a content has something that should not be shown on a conversation. On this course, we use TensorFlow.js (machine learning in JavaScript), a pretrained model, for content moderation. It is very easy to use: just call the model and get a reply, a list of types of contents to be blocked and their respective flags.Online Content moderation is the process by which one must decide what stays online vs. what should be moderated or even eliminated; no one is happy to be moderated, I was not when I was the target of several online moderations on YouTube, Facebook and Twitter. The "Napalm girl" was a famous case in which Facebook classified as nudity, which is a historical picture.What is the best and most efficient way to safely moderate content online?Several programming savvies are doing their best. Like in String Theory, they gather the best, but no practical answer came out yet. Recently, we had chatGPT that can make sentiment analysis in an advance approach, this course shows an alternative, which is free. chatGPT has a small chart per API call. Furthermore, chatGPT requires much less coding expertiseOn this course, I want to share a trivial case, which was not also easy to solve for a scientists, but now easily implemented as public library: direct insult (i.e., calling people names).The bot is a classical pattern recognition: it gives out a number, and you must decide your tolerance for insults. This number, already standard on deep learning, is a sort of probability, loosely speaking.We are going to use two MIT license Google public tools: TensorFlow.js and Angular. The former is a deep learning tool for freely creating smart web application, whereas the latter is a free framework for creating frontend application, no need to create servers.
