| 성과구성 |
Abstract
This paper descnl>es an approach to extractinga predicate-argumentstructure (PAS) in building design rule sentences using natural language processing (NLP）and deep loaming models. For the computer to reason about the compliance or buildingdesign, design rules represented by natural language mustbe converted into a computer-readable rormaLThe rule
interpretation and translation processes are allenging tasksbecauseof the vagueness and ambiguity of natural language Many studies have proposed approa es to address this problem, but most or these are dependent on manual tasks, which is the bottlen to expanding the scope of design rulechecking to design requirements from various documents. In this paper, we applydeep learning-bas려 NLP te야 niques for translating design rule sentences into a computer-readabledata structure. To apply deep learning-based NtP techniques to the rule interpretation process, we identi.fied the semantic role
elements of building design requirements and defined a PAS for design rule checking. Usinga bidirectional longshort-term memory model with a conditional random field layer, the computer can intelligently analyze constituents of building design rule sentences and automatically extract tlie logic:al elements. The propos approach contributes to broadening the scope of buildinginfomiation modeling-enabled rule checking to any nacural language-ba.s design requirements.
Keywords: automated rule ecking;building information modeling(BJM); natural language processing(NLP);predi cate argument structure |
