Modeling and analyzing dynamic social networks for behavioral pattern discovery in collaborative design

Abstract

As a newly-developed information exchange and management platform, Building Information Modeling (BIM) is altering the way of collaboration among multi-engineers for civil engineering projects. During the BIM implementation, a large number of event logs are automatically generated and accumulated to record details of the model evolution. For knowledge discovery from huge logs, a novel BIM event log mining approach based on the dynamic social network analysis is presented to examine designers’ performance objectively, which has been verified in BIM event logs about an ongoing year-long design project. Relying on meaningful information extracted from time-stamped logs, networks on the monthly interval are built to graphically represent information and knowledge sharing among designers. Special emphasis is put on measuring designers’ influence by a defined new metric called “impact score”, which combines the k-shell method and 1-step neighbors to achieve comparatively low computational cost and high accurate ranking. Besides, an emerging machine learning algorithm named CatBoost is utilized to predict designers’ influence intelligently by learning features from both network structure and human behavior. It has been found that twelve networks can be easily distinguished into two collaborative patterns, whose characteristics in both network structures and designers’ behaviors are significantly different. The most influential designers are similar within the same group but varied from different groups. Extensive analytical results confirm that the method can potentially serve as month-by-month feedback to monitor the complex modeling process, which further supports managers to realize data-driven decision making for better leadership and work plan towards an optimized collaborative design.