Good point. I assumed relevance was approximately similar to the correlation, without a strong assumption on the signs. Which is just my interpretation, not a universal definition.