Construction cost estimation is a labor-intensive task that involves several processes. Although some of these processes have been automated, construction cost estimation still relies heavily on manual inputs. To compute a cost estimate, an estimator needs to: (1) take off quantities and extract some required cost information from the architectural model or drawing; (2) extract other required cost information from the construction specifications; (3) assign building elements to work or cost items; and (4) retrieve the unit cost of the work or cost items to further compute the cost estimate. To achieve full automation of construction cost estimation, the manual inputs required to classify building elements, and to retrieve pricing information of work items need to be automated. To address that, the authors proposed a new method that uses semantic modeling and natural language processing techniques in developing algorithms that automate the manual processes involved in: (1) extracting design information from construction specifications; (2) using the extracted information to match specified material in the construction specifications with items from an established database; and (3) retrieving the pricing information of the materials specified in the construction specifications. To test the validity of the authors' proposed method, an experiment was conducted using eight wood construction projects in Detroit, MI. The proposed method was utilized to develop an algorithm that can process the construction specifications automatically and retrieve the unit cost of materials from a database. The results from the developed algorithm were compared with the gold standard (results manually generated by industry experts). The developed algorithms achieved 99.2% precision and 99.2% recall (i.e., 99.2% F1-measure) for extracted design information instances; 100% precision and 96.5% recall (i.e., 98.2% F1-measure) for extracted materials from the database. The authors demonstrated that as the training data increases, the performance levels increase. The developed algorithms utilized 5.56% of the time it took using the current traditional method of extracting design information from construction specifications manually. These results showed that the proposed method is promising in developing algorithms that automate the processing of construction specifications to extract the design related information in fulfilling essential information requirements of detailed wood construction cost estimation and in retrieving the unit costs.

建筑成本估算是一项劳动密集型任务，涉及多个过程。尽管其中一些流程已实现自动化，但建筑成本估算仍严重依赖于人工输入。为了计算成本估算，估算者需要： (1) 计算数量并从建筑模型或图纸中提取一些所需的成本信息；(2) 从施工规范中提取其他所需的造价信息；(3) 将建筑元素分配给工作或成本项目；(4) 检索工作或成本项目的单位成本以进一步计算成本估算。为了实现建筑成本估算的完全自动化，对建筑元素进行分类和检索工作项目的定价信息所需的手动输入需要自动化。为了解决这个问题，作者提出了一种新方法，该方法使用语义建模和自然语言处理技术来开发自动化手动过程的算法：（1）从施工规范中提取设计信息；(2) 使用提取的信息将施工规范中的指定材料与已建立的数据库中的项目相匹配；（三）检索施工规范规定的材料价格信息。为了测试作者提出的方法的有效性，在密歇根州底特律使用八个木结构项目进行了一项实验。所提出的方法用于开发一种算法，该算法可以自动处理施工规范并从数据库中检索材料的单位成本。将开发算法的结果与黄金标准（由行业专家手动生成的结果）进行比较。所开发的算法对提取的设计信息实例实现了 99.2% 的准确率和 99.2% 的召回率（即 99.2% F1-measure）；从数据库中提取的材料具有 100% 的精确度和 96.5% 的召回率（即 98.2% F1-measure）。作者证明，随着训练数据的增加，性能水平也会提高。使用当前手动从施工规范中提取设计信息的传统方法，开发的算法使用了 5.56% 的时间。