XiOs, I work along side Wildhoney so I feel obliged to comment. I can see absolutely no advantage from doing this relatively straight forward task your way, you seem only to make this task more difficult to implement, which in turn will take longer and cost the client more money. If you're some sort of cowboy then that's the way forward, but we're not. While parsing the HTML documents, you can add weight to specific keywords and key phrases, depending on the tags which encapsulate them.

The solution offered by Wildhoney is practical, Why over-complicate a task? Admittedly parsing HTML documents is not the most efficient solution, but when you weigh the amount of processing required for parsing HTML versus the frequency of processing (i.e. in this situation processing will only be required every x hours/days/weeks/etc.) then the slight performance loss from using this method is irrelevant.

And afterall, this method is fine for companies such as Google, so why not for this situation?