Development of energy codes requires an appropriate building stock representation. To this end, information available within municipal construction permit records (e.g., dwelling type, floor and footprint area, foundation type, number of above-grade storeys, availability of an attached garage, and number of bedrooms) represents an untapped opportunity. However, these records are often stored in a semi-structured fashion with relevant information scattered within free-form text description fields, which necessitates text analytics. This paper introduces a text analytics-based method, employing the association rule mining, for information retrieval from permit records to support building stock modelling process; and the method is demonstrated with permit records from small residential buildings in seven major Canadian cities comprising over 240,000 unique entries. The analysis highlights that detached housing is the most common housing type and the average dwelling floor area was between 174 and 266 m². Each permit record is associated with existing Canadian housing archetypes. It is found that mid-to-large-size detached archetypes are more common than smaller row or semi-detached archetypes. A building performance simulation-based investigation revealed that assuming an equal weight for each archetype, instead of using the distributions identified from the permit records, causes an underestimation of the housing stock's energy use by 36%.