營建產業擁有高風險之特性,並存在許多足以導致重大傷亡之潛在危險,因此各國的營建業都有著降低與消除營建工安事故的機制存在。「作業危害分析」是美國營建業常見的營建工安自主檢查,其透過事前分析營建專案當中的工程作業項目,辨識其背後的潛在危險因子,並針對潛在危險提出預防措施。本研究基於「協助作業危害分析」的目的建立營建工程安全領域的知識本體(Ontology),但與既往之研究相較,選用更普遍及易得的文字來源。本研究首先蒐集三種可能在作業危害分析過程當中作為參考的異質性文件:(1)描述危險情境與預防措施的「作業危害分析文件」(2)描述重大傷亡案例的「重大傷亡檢討報告」(3)描述工地所應遵循之規範標準的「工程安全規範」。在取得三種文字資源後,再利用同質性高的「作業危害分析文件」當成機器學習之基礎,應用文件自動分類技術發展出最佳之分類策略,接著再將文件自動分類策略配合資訊檢索技術輔助,推廣至異質性的「重大傷亡檢討報告」,而逐步完成營建工程安全領域之知識本體。研究發現,針對某一特定類別所發展之分類策略雖然在套用至異質性文件時無法全然達到預期之成效,但除了對整合異質性文件之策略提出了方向與建議之外,在進行同質性之「作業危害分析文件」的自動分類上則有得到良好的分類成效,此部份對於「協助作業危害分析之自動化流程」上已有一定程度的幫助。
Construction industry has higher potential on occupational hazard than other industries do. To prevent from the fatalities and injuries occurred in construction project, Job Hazard Analysis (JHA) is a possible approach. It identifies all the activities in a construction project, recognizes the potential hazards behind each activity, then recommends possible safety approaches to eliminate the potential hazards. In order to assist JHA, this research proposes a semi-automated approach to develop a construction safety domain ontology which is based on Information Retrieval (IR) and automatic document classification techniques. Different from similar research, this research adopts more general text resources to develop the ontology. In the first step, this research collects three different types of documents which can provide references to JHA. The three types of construction safety documents are: (1) JHA documents which contains activities, hazards and safety approaches (2) fatality case reports (3) construction safety standards. In the second step, this research performs Machine Learning techniques over the JHA documents to find the best strategies for optimizing the effectiveness of automatic document classification. In the third step, the strategies are combined with Information Retrieval (IR) techniques and then applied to the automatic classification of fatality case reports. By these procedures, this research shows how to develop the construction safety domain ontology step by step. The conclusion is that although the effectiveness of integrating the different types of construction safety documents still has room for improvement, this research discusses the possible reasons behind the insufficient effectiveness and also provides several suggestions to improve the effectiveness. Moreover, the document classifying strategies this research suggests still achieve good effectiveness within JHA documents, meaning that it still has contributions to Job Hazard Analysis.