Harassment can be flagged up either by the type language used or the repetition of an action. It has to start with a manual we-know-it-when-we-see-it process.

Harassment can be flagged up either by the type language used or the repetition of an action. It has to start with a manual we-know-it-when-we-see-it process. You collect data about the types of users that are using your service in a harassing way. Where they come from, how they use the service etc. Over time you will see patterns and trends that will likely lead to you being able to flag certain sign ups in that way. The more accurate it gets over time the less manual intervention you will need. It maybe that certain referrers or keywords drive this type of traffic. Perhaps there are similarities in the first actions of this type of new user or the content that they create. The only way to recognise it is to start gathering examples and learning from them.