What are AI Classifiers?

Lately, we are all hearing about AI non-stop. Similarly, many of us may not realise how much we interact with AI daily be that from search engines, ads, social media and more. Despite all the recent AI talk many of us aren’t aware of how it works, specifically in relation to content moderation and AI classifiers.

An AI classifier is a machine learning model trained to analyse and categorise content. This can be in the form of images, videos, text, and audio. This is done based on predefined criteria. When it comes to moderation these classifiers’ goal is to be able to identify and flag content that violates platforms guidelines and the law.

Some examples of classifiers are:

·         Weapons. Such as guns, bombs & more

·         Hate speech

·         Spam

·         Terrorism

·         Narcotics

·         Extremism

·         Sexism

·         Nudity

·         NSFW

·         Racism

·         Violence

·         War Crime

·         CSAM

The list goes on. How classifiers are trained can vary depending on the company. It usually consists of training data, feature extraction, model training, validation and fine-tuning, deployment and retraining. Certain types of classifiers are easier to train than others. Indeed, training a classifier to identify CSAM is one of the most challenging classifiers one can build due to the nature of the content. Orthus/T3K has one of the best on market CSAM classifiers and has been trained in partnership with law enforcement. On top of this the way that we approach training our classifiers is what makes them have one of the highest true positive rates on the market. These can also be adapted and customised to platforms specific moderation needs.