I realize that Google writes all the contents within the page and use css 'id' or 'class' to distinguish program elements of different API levels.
I solve this problem by using html parser, such as beautiful soup or lxml, to look for (select) specific css tags, that work pretty well.