Selecting actions for resource-bounded information extraction using reinforcement learning