Modern buildings integrate a multitude of interconnected sensors and control devices for efficient energy management. Different conventions are adopted by vendors, technicians, and operators for naming and creating metadata for these sensors and devices. Analytical tools for understanding system status, fault diagnosis, and optimizing energy consumption in such complex environments necessitate automated categorization or classification of sensors and their metadata according to a standard ontology like Brick or Haystack. Previous research has predominantly focused on automatically inferring only the classes of the sensors from their text name strings. However, there exists an evident void in research concerning the automated inference of sensor metadata (i.e., properties) and in this paper we aim to address this novel and challenging problem. Our contribution lies in developing data-driven approaches that combine NLP based context aware representation of sensor names with machine learning to accurately infer sensor properties. The effectiveness of the developed approaches is assessed using a proprietary dataset comprising 129 buildings. The results of this study hold implications for improving the portability and adaptability of energy analytics applications within smart building systems.