๐ Welcome To The Friendly Text Moderation
Identify 14 categories of text toxicity.
This NLP (Natural Language Processing) AI demonstration aims to prevent profanity, vulgarity, hate speech, violence, sexism, and other offensive language. It is not an act of censorship, as the final UI (User Interface) will give the reader, but not a young reader, the option to click on a label to read the toxic message. The goal is to create a safer and more respectful environment for you, your colleages, and your family. This NLP app is 1 of 3 hands-on courses, "AI Solution Architect," from ELVTR and Duc Haba.
๐ด Helpful Instruction:
Enter your [harmful] message in the input box.
Click the "Measure 14 Toxicity" button.
View the result on the Donut plot.
(Optional) Click on the "Fetch Real World Toxic Dataset" below.
There are additional options and notes below.
๐ฅ WARNING: WARNING:
- The following button will retrieve real-world offensive posts from Twitter and customer reviews from consumer companies.
- The button will display four toxic messages at a time. Click again for four more randomly selected postings/tweets.
- They contain profanity, vulgarity, hate, violence, sexism, and other offensive language.
- After you fetch the toxic messages, Click on the "Measure 14 Toxicity" button.
๐ป Author and Developer Notes:
The demo uses the cutting-edge (2024) AI Natural Language Processing (NLP) model from OpenAI.
This NLP app is 1 of 3 hands-on apps from the "AI Solution Architect," from ELVTR and Duc Haba.
It is not a Generative (GenAI) model, such as Google Gemini or GPT-4.
The NLP understands the message context, nuance, innuendo, and not just swear words.
We challenge you to trick it, i.e., write a toxic tweet or post, but our AI thinks it is safe. If you win, please send us your message.
The 14 toxicity categories are as follows:
- harassment
- harassment threatening
- harassment instructions
- hate
- hate threatening
- hate instructions
- self harm
- self harm instructions
- self harm intent
- self harm minor
- sexual
- sexual minors
- violence
- violence graphic
If the NLP model classifies the message as "safe," you can still limit the level of toxicity by using the "Personal Safe" slider.
The smaller the personal-safe value, the stricter the limitation. It means that if you're a young or sensitive adult, you should choose a lower personal-safe value, less than 0.02, to ensure you're not exposed to harmful content.
The color of the donut plot is as follows:
- Red is an "unsafe" message by the NLP model
- Green is a "safe" message
- Yellow is an "unsafe" message by your toxicity level
The "confidence" score refers to the confidence level in detecting a particular type of toxicity among the 14 tracked types. For instance, if the confidence score is 90%, it indicates a 90% chance that the toxicity detected is of that particular type. In comparison, the remaining 13 toxicities collectively have a 10% chance of being the detected toxicity. Conversely, if the confidence score is 3%, it could indicate any toxicity. It's worth noting that the Red, Green, or Yellow safety levels do not influence the confidence score.
The real-world dataset is from the Jigsaw Rate Severity of Toxic Comments on Kaggle. It has 30,108 records.
- Citation:
- Ian Kivlichan, Jeffrey Sorensen, Lucas Dixon, Lucy Vasserman, Meghan Graham, Tin Acosta, Walter Reade. (2021). Jigsaw Rate Severity of Toxic Comments . Kaggle. https://kaggle.com/competitions/jigsaw-toxic-severity-rating
The intent is to share with Duc's friends and colleagues, but for those with nefarious intent, this Text Moderation model is governed by the GNU 3.0 License: https://www.gnu.org/licenses/gpl-3.0.en.html
Author: Copyright (C), 2024 Duc Haba
๐ "AI Solution Architect" Course by ELVTR
Welcome to the fascinating world of AI and natural language processing (NLP). This NLP model is a part of one of three hands-on application. In our journey together, we will explore the AI Solution Architect course, meticulously crafted by ELVTR in collaboration with Duc Haba. This course is intended to serve as your gateway into the dynamic and constantly evolving field of AI Solution Architect, providing you with a comprehensive understanding of its complexities and applications.
An AI Solution Architect (AISA) is a mastermind who possesses a deep understanding of the complex technicalities of AI and knows how to creatively integrate them into real-world solutions. They bridge the gap between theoretical AI models and practical, effective applications. AISA works as a strategist to design AI systems that align with business objectives and technical requirements. They delve into algorithms, data structures, and computational theories to translate them into tangible, impactful AI solutions that have the potential to revolutionize industries.
๐ Sign up for the course today, and I will see you in class.
- An article about this NLP Text Moderation will be coming soon.