Contents lists available at ScienceDirect Automation in Construction journal homepage: www.elsevier.com/locate/autcon Project collective mind: Unlocking project discussion networks Mazdak Nik-Bakht a,⁎ , Tamer El-Diraby b a Department of Building, Civil & Environmental Engineering, Concordia University, Montreal, Quebec, Canada b Department of Civil Engineering, University of Toronto, Toronto, Ontario, Canada ARTICLE INFO Keywords: Social network analysis Twitter Semantic analysis Project management and monitoring Stakeholder mapping Community engagement ABSTRACT A project discussion network is a space where project stakeholders form relationships among each other and share information about the project. Virtual discussion networks may refer to networks of e-mails, document exchange and social media (such as Twitter, Facebook, YouTube, etc.). As such, both social linkages and se- mantics of the exchanged content must be considered in analysis of such networks. The proposed framework in this study aims to analyze both the social and semantic aspects of these networks. We developed the framework through analysis of the social networks formed around Twitter accounts of infrastructure megaprojects. To assure relevance to construction research and practices, three objectives guided our analysis: relaying on a large and diversified data corpus from construction projects; testing the applicability and usage of a set of relevant algorithms to the context of construction project management; and linking the results of data analysis and algorithm evaluation to the conditions of construction projects at hand. In examining algorithms for detecting sub-communities, the Louvain fast unfolding modularity maximization was more suitable in detecting project relevant sub-groups. For assessing the relative influence of actors, PageRank algorithm performed better than centrality measures. For extracting key terms, we found that modifying the term frequency-inverse document frequency (TF-IDF) measure to incorporate the relative importance of the source nodes enhances the relevance of extracted terms. Obliviously, Twitter networks are only one type of project networks that can cover a limited/ biased sample of participants. Their analysis should be one component of the overall project network analysis. We believe that the proposed framework has the same level of applicability to internal networks of project teams as well as non-Twitter networks. 1. Introduction Projects are essentially networked phenomena. People interact and build complex relationships with others who have interests in a project throughout the project life cycle. In the early stages of applying social network analysis (SNA) in construction, researchers focused on ana- lyzing case studies to explore actual social ties between project players (see for example: [28]). With the transformation of computers from automated calculators to communication amplifiers [24], interest started to grow in the use of SNA to study communication patterns and their relationships to project organization [11]. Project “networked” stakeholders exchange a large volume of information—in the form of documents, e-mails, and even contribution to the project social media outlets. In other words, a project embodies (or can be perceived as) a social network of knower agents (people), overlaid by (semantic) net- works of exchanged content encapsulating not only their views but also their knowledge and experience. If we analyze both networks, we can extract insightful measures in relation to the relative influence of project players, flow of information and its impact on the opinion dy- namics, etc. We can also detect major problem areas and distill ideas for possible solutions. Repeated analysis can reveal patterns of concept association, which can be used to enrich our knowledge models. Until now, limited work has been directed towards integrated so- cial-semantic analysis of project networks. Further, the use of SNA in construction has been limited by the size of project networks due to the difficulty of tracking communication between project professional staff. In most cases, researchers had to construct the social network through interviews and observation (see for example: [2,38]). The emergence of social media usage in construction projects can be a helpful in auto- matically capturing large sets of data over a long period of time. Moreover, analysis of social media outlets related to construction pro- jects can be helpful in better managing project relationships with its community [36]. However, current work on analysis of project social media accounts is lagging [19]. In this paper, we advocate that socio-semantic analysis of project discussion [social] networks (PDN) should be a major task for decision http://dx.doi.org/10.1016/j.autcon.2017.08.026 Received 15 November 2016; Received in revised form 3 May 2017; Accepted 11 August 2017 ⁎ Corresponding author. E-mail addresses: mazdak.nikbakht@concordia.ca (M. Nik-Bakht), tamer@ecf.utoronto.ca (T. El-Diraby). Automation in Construction 84 (2017) 50–69 0926-5805/ © 2017 Elsevier B.V. All rights reserved. MARK