How accurate are the moderation results? What does the confidence score mean?

Accuracy is one of our key priorities. Here is how we achieve top-level accuracy:

• We use state-of-the-art technology and proprietary algorithms for our moderation.
• We have been working with customers and learning from them since 2013. This experience helps us achieve excellent results in most circumstances.
• We perform updates and improve the moderation engine on a continuous basis.
• We constantly monitor and regularly benchmark our accuracy levels.
• We have a Feedback API available to users who wish to share their target classifications to help us get even better.

Actual accuracy metrics will vary depending on the characteristics of your datasets, but as a general guideline we have found our Nudity Detection to have a 99.2% F1-score on porn detection.

Along with each moderation result you will get a confidence score. The confidence score is a number between 0 and 1. The closer it is to 1, the more confident the model is. See this question to know more about how to interpret the scores returned by the API and how they can help you decide on actions to take.