Interpreting the similarity score