I'm a big fan of using XML::LibXML in HTML-parsing mode along with some good XPath knowledge to identify elements to extract and alter. True, there's some really broken HTML out there that doesn't work, but for most stuff, it's a good tool. I've written a couple of columns about that.