Although it is impossible to censor and scrutinize every user's utterance that is out in the digital space, it is helpful to utilize moderation tools that can diminish offensive online behavior.
Some of the most common tools that assist with these efforts are:
- Google Perspective API
- Microsoft Azure Content Moderator
- Two Hat (Community Sift)
- Hive Moderation
- Meta's AI Moderation
Each of these tools is tailored to specific platforms with their unique needs. For example, Microsoft Azure can detect profanity, adult content, or personal information in social platforms and enterprise-level applications. It is also good at scanning images and videos, and not just simple text. Two Hat is good at real-time chat moderation and is suitable for gaming and youth communities. And Google Perspective API supposedly identifies toxicity in online comments.
Google Perspective is interesting because it is trained both by Labeled Human Data and Machine Learning Techniques such as Natural Language Processing, User Feedback, and Supervised Machine Learning. Labeled Human Data is a collective effort in which the word labeling is done by a varied group of people and experts to ensure low bias rates. They determine the levels of toxicity of a word and assign a particular category to it, such as: toxic, high level of toxicity, hate speech, identity attack, somewhat toxic, etc. This labeling isn't arbitrary but is powered by socio-cultural context and nuances, tone, and intent.
No comments:
Post a Comment