I think it would be useful if the module scanned all nodes before saving a keyword so that it took into account its number of appearances throughout all the content. In my opinion, whether or not one word is mentioned two or three times in a node, doesn't tell much about the cloud we want to build in the end. (current functionality)