How could I go about making a context-based text moderation powered by a machine learning algorithm?

I need to create a machine learning algorithm that can predict weather or not a sentence is inappropriate. It needs to be fast and has to be context based, so words will only be blocked in certain scenarios. An example: "Address" would not be blocked, but "What is your address" would be.

My plan was to have a database filled with all the blacklisted words, and then inside those documents in the database, I would have an array of all the different scenarios that it would be blocked. I would then use a machine learning algorithm to predict how similar a piece of inputted text is to any of the scenarios.

I think that that would work fine, but it would be incredibly time consuming to create without a massive team to help block words (I am on my own). Would there be a better way to do this?