Putting Similarity Assessments into Context: Matching Functions with the User's Intended Operations