A clustering approach for analyzing the computability of building code requirements

Automated code compliance checking has been implemented using a variety of different methods. These methods vary widely in terms of automation level, expressivity, representativeness, accuracy, and efficiency. Although most of the existing methods/tools are similar in that they represent the code requirements in the form of rules, the semantic representation and language/format they use for rule representation vary from a method/tool to another. One common limitation of all these methods/tools is the inability to fully represent all types of requirements (e.g., requirements that are too complex or that require human judgment by nature). More research is, thus, needed to better identify what the different types of requirements are - in terms of their semantics and syntactics - and which of them could be represented in a computer-processable form. To address this need, this paper proposes a new methodology for identifying the different types of building code requirements in terms of computability. The proposed methodology consists of three primary tasks: (1) use a machine-learning clustering algorithm to identify the computational clusters of the code requirements, (2) identify the different types of requirements in the different clusters, in terms of their semantic and syntactic features and structures, complexity, etc., and (3) determine the computability of different sentence types. The proposed methodology was implemented and tested on a corpus of sentences from the 2015 International Building Code. The paper discusses the clustering and analysis methodology and results.