Management of construction safety and health plans based on automated content analysis

Abstract

Safety management in the construction sector continues to be of great concern due to the high accident rate. Enormous legislative efforts have been made to minimize this problem. In Europe, the Directive 92/57/EEC, which establishes a mandatory safety and health requirements, defines the Safety and Health Plan. This paper contains a methodology for automatically extracting structured relevant information from this textual document. Such plans are crucial to provide safety in this sector. However, current documents are often enlarged with nonessential information, while the most relevant information is missing or difficult to locate. The information extracted by our methodology makes it easier for Safety and Health Plans to be validated and approved at an early stage, as well as facilitating their later use and updating. The main steps of this methodology are: i) the definition of the relevant content to be extracted as seven items, in collaboration with experts in the area; ii) text preprocessing and enrichment using existing software resources for the natural language employed; and iii) automated content analysis for extracting information, on the basis of a collection of previously established rules for identifying the content defined in the first step. The output is a detailed and structured report containing the percentage of relevant information identified, as well as the desired items of information. To illustrate and validate our methodology, we have implemented and applied it to 50 real Safety and Health Plans. As a result, our system was able to identify most of the required information for five of the seven items in all of these plans. In addition, we have been able to detect the problems in the textual plan that caused the missing information in the other two items, which will be corrected in future developments.