Currently, we're only returning nodes based on the first category listed in the mappings.yml file, therefore the ordering of these categories is important. I believe the node normalizer returns node categories in order of specificity, with the most specific category first (which is what we want), so we might want to look into calling node norm to get the list of categories.
This all might be a moot point if we reimplement the precomputed correlations that Patrick had started work on.