Document and Query Expansion for Information Retrieval on Building Regulations

Regulations and test criteria for building products are captured in hundreds of interrelated documents. It can be daunting to figure out which of these documents contain information that is relevant to your building project or product. In this paper, we describe work on an Information Retrieval (IR) system that aims to search through the contents of building regulations. Based on practitioner interviews we develop a small dataset of user-queries for which we would like to return relevant passages of documents. We explore several approaches to Query Expansion (QE) and Document Expansion (DE), taking into account the scarcity of openly available knowledge sources in our small technical domain. We show that IR performance can be greatly improved using QE and DE, and retrieve a top-3 relevant result for up to 85% of out queries. We share our IR dataset and the code to replicate our approach.