As the pages were generated programmatically, I think you would be more successful scraping the relevent data out of the surrounding markup. This avoids traversing the DOM or mucking around with regular expressions. See

Template::Extract
Text::Scraper