Do the same thing that most wikis use and create a new notation with well defined rules. Then convert that to HTML. Assuming your new notation is well designed that would be the end of the problem. HTML is not suited to this as it has too many edge cases where things can happen that you may not anticipate, and filtering those constructs out is difficult if you intend to allow some of them to pass through, requiring you to handle parsing html, which is expensive.

Also if go with HTML-like markup you need to consider carefully where you handle task like filtering. Putting on the submit has different ramifications from putting at the fetch.