Cyclone3 Profanity Filter

Basic principles

The filter learns automatically. It tries to understand the provided text, its semantics, special character, link content and other properties of the text. It uses standard algorithms for conditional probability together with its own algorithms. It can analyze standard articles on the server and identify off-topic text as spam. We also remember infamous authors who repeatedly submit unacceptable messages and consider this fact in the text analysis by utilizing a system of karma.

The reliability is closely related to the volume of training data on your site. We also have our own database of training data collected from various sites.

Simple API

Cyclone3 Profanity Filter is a remote service running on our servers. You can implement it regardless of the technology and programming language you use. Simple SOAP, XML and HTTP are used for communication.

Usually, the basic attributes of a message are:

• message text (content)
• author - unique identifier (optional)
• your own arbitrary scoring of the text (optional)
  

Learning properties

If you provide text scoring in the request, the filter will consider it as training data. It is entered into the training database and processed. This way you can correct the system if it provides unsatisfactory answers.