This document proposes a framework to automatically manage domain and range information for knowledge entries in a knowledge base. It does this by using word embeddings to generate feature vectors for subjects and objects of relationships, and then training supervised machine learning models to classify whether elements belong to the domain or range of each relationship. The models are generated from positive and negative training examples of existing knowledge entries. An experiment tests the accuracy of the models on 32 common relationships, showing they can accurately detect correct and incorrect entries.