The thing about building automated trading systems (related or not to Semantic analysis) is that the things you tend to track are compared to other things you are tracking. After all, if there is an issue with a data stream you wouldn't want to bet the farm on it. You check with other streams. So goofy stories on april first wouldn't affect the algorithms too much unless they were reported the same way in disparate news/feed sources.