Clustering-Based Approach for Building Code Computability Analysis

One common limitation of all automated code compliance-checking methods and tools is their inability to deal with all types of building-code requirements. More research is needed to better identify the different types of requirements, in terms of their syntactic and semantic structures and complexities, to gain more insights about the capabilities and limitations of existing methods and tools (i.e., which requirements they can automatically process, represent, or check, and which not). To address this need, this paper proposes a new set of syntactic and semantic features and complexity and computability metrics for code computability analysis. A clustering-based approach was used to identify the different types of code sentences, in terms of their computability, using the proposed features and metrics. The approach was implemented and tested on a corpus of 6,608 sentences from the International Building Code and its amendments. The sentence clusters and identified sentence types were evaluated using intrinsic and extrinsic evaluation methods. The evaluation results indicated good clustering performance, perfect alignment between the human- and computer-identified types, and good agreement in the assignment of sentences to the types.