Batch recent replies and send to AWS Comprehend for content moderation

[Arizard]

My idea is that we periodically send a batch of recent replies to AWS Comprehend for content moderation. Flagged replies are marked as such in db and then censored when returned from /replies.

This prevents leaking PII and hosting abusive text content.

[Arizard]

See here: https://aws.amazon.com/blogs/machine-learning/flag-harmful-content-using-amazon-comprehend-toxicity-detection/#:~:text=The%20Amazon%20Comprehend%20Toxicity%20Detection%20API%20assigns%20an%20overall%20toxicity%20score%20to%20text%20content%2C%20ranging%20from%200%E2%80%931%2C%20indicating%20the%20likelihood%20of%20it%20being%20toxic.

[Arizard]

Custom model training can be used to detect PII, Toxicity and Spam. Might be a bit complex to achieve but kinda cool.