Process Analysis With an Automatic Mapping of Performance Factors Using Natural Language Processing

In lean construction projects, much information is collected during the process analysis with the trades. This data is increasingly documented as a reference for use in future construction projects. By doing this, efficient methods are required to use this data. Often, the unstructured naming of data is a challenge for a rule-based allocation of information, and manual work is required to identify the needed data. Therefore, the aim is to develop an automatic mapping of historical performance factors to the tender specifications of a new construction project. To support the process analysis with historical project data, a case study is executed using Natural Language Processing (NLP). With a NLP model, the process descriptions from the tender specifications of the new construction project can be compared with a master database, to filter the right performance factor and calculate the duration for a process. This procedure can be used to support the further process analysis together with the trades to generate a validated construction schedule. The case study shows promising results in the prediction results. First, the mapping quality and second, the prediction accuracy are evaluated. With the developed mapping concept, last planners can validate their estimations of durations in lean construction process planning with a target to support stability in a project. Still, a more detailed description of the processes could increase the prediction results.     «

In lean construction projects, much information is collected during the process analysis with the trades. This data is increasingly documented as a reference for use in future construction projects. By doing this, efficient methods are required to use this data. Often, the unstructured naming of data is a challenge for a rule-based allocation of information, and manual work is required to identify the needed data. Therefore, the aim is to develop an automatic mapping of historical performance fa...     »