Knowledge Management Framework of Construction Accident Cases Using Natural Language Processing

Abstract

Construction accident cases include knowledge to identify risk in similar situations and to establish safety measures. For this reason, knowledge management of construction accident cases is important because it can prevent accidents by controlling risks on site. Accordingly, a lot of research has been conducted to manage knowledge of construction accident cases. However, since accident cases are recorded as unstructured text data there are limitations, requiring significant time and effort to retrieve and analyze the knowledge the user wants. To overcome these limitations, This research proposes a framework of knowledge management system for construction accident cases using two NLP technologies: IR and IE. In the Semantic Retrieval Model using IR, the query was expanded by establishing a thesaurus that integrates the unique expressions used in accident cases and common terms in the general construction industry. The ranking of the retrieval results was calculated considering the Okapi BM25 and the weighting according to the semantic level of the thesaurus. In the Tacit Knowledge Extraction Model, tacit knowledge was automatically extracted from each accident case retrieved through rule-based and machine learning (CRF), statistical analysis was performed, and the analysis results were visualized. The prototype system was developed using Python to implement the proposed methodology. The proposed system can retrieve results that are 97% relevant to the accident cases the user intended, and automatically analyzed knowledge with an accuracy measure of 93.75% and 84.13% for the rule-based and CRF models respectively. The evaluation results show that the system has the ability to retrieve for similar accident cases intended by the user and to automatically extract available knowledge from the accident cases. This research has enabled the effective use of knowledge necessary to prevent accidents by managing accident case knowledge through an automated retrieval and analysis system. In addition, this research was intended to provide a basis for knowledge management that can respond to uncertainties related to the safety of construction sites by promptly supporting decision making related to construction safety management. This research look forward to enhance the capacity of construction safety management.