The semantic web is a solution in search of a problem; most websites don't have semantics beyond "the page looks like this", and couldn't be trusted to describe them accurately if they did. Enabling a machine to understand a website in the same way that a human can would be valuable, but is probably an AI-complete problem; in the meantime the best path towards it is probably platforms for easy mostly-automated web scraping or testing (In fact the most useful stepping stone towards a semantic web I've seen is probably adblock, with its simple slider/checkbox interface for telling it "this piece is an ad" in the cases where it failed to figure that out).
Maybe "why are Selenium tests so flaky? Can we do better?" would be a good starting point if you want to work on this problem. The inability of most (all?) companies to write good browser-based test suites is effectively those companies' inability to recover the semantics of their own site.
I never made the link between semantic web and the pain that are browser tests, but now that you say it, it makes sense (especially if you use semantic attributes to identify the page elements for your tests).
Maybe "why are Selenium tests so flaky? Can we do better?" would be a good starting point if you want to work on this problem. The inability of most (all?) companies to write good browser-based test suites is effectively those companies' inability to recover the semantics of their own site.