AI Approaches to a Question Answering System for a National Code

Strict code compliance processes require accurate and correct information, so when searching this significant information base the identified results must present factual information and exact answers to what is being searched for. Incorrect or inaccurate information can cause failure to comply, resulting in a loss of time, effort, and money. Question answering (QA), a natural language processing (NLP) task, can enhance efficiency by automatically retrieving relevant information. BERT, a high-performance deep learning model, is commonly used in QA systems for understanding human questions and language, and retrieving accurate answers to questions that are input. BERT has been shown to excel within the construction domain in extracting answers from textual construction documents. RoBERTa, its successor, claims to offer superior performance through a more significant training corpus, longer input sequences, and a refined pre-training approach. This research aims to investigate the impact of a RoBERTa based QA system and compare its performance to BERT in QA for building codes.