# Automating Vocabulary Mapping In most cases, a manual mapping was created for each collection x CV combination. Can these manual mappings be effectively replaced by machine learning? We can determine this using the existing manually coded mappings to easily provide different sized traning sets for each collection and each controlled vocabulary. For each combination of collection and controlled vocabulary: 1. Which fields provide input useful for determining the vocabulary term? 2. What is the response of precision and recall to the size of the training set?