Identification of Metrics for the Purdue Index for Construction Using Latent Dirichlet Allocation

Abstract

The construction industry is one of the most significant contributors to the growth of the US economy as well as the global market. The Purdue Index for Construction (Pi-C) was developed in the form of a composite index consisting of five dimensions (Economy, Stability, Social, Development, and Quality) to monitor the health status of the construction industry and facilitate data-driven decision making. Despite its great potential, metrics under the Development and Quality dimensions are still missing, which limits our understanding of the health status of the construction industry. A promising approach to identify the missing metrics is to apply the latent Dirichlet allocation (LDA), which supports the discovery of latent topics from a large set of textual data. In this regard, this work introduces an LDA-based method to identify new metrics for the Development and Quality dimensions of the Pi-C. A total of 10,466 abstracts of research papers relevant to Development and Quality were collected from academic search engines using a web crawler. The LDA analysis was conducted to identify metrics and corresponding variables. As a result, two new metrics—Technology and Education—in the Development dimension and one new metric—Sustainability—in the Quality dimension were identified for Pi-C. Results revealed that the updated Pi-C improves our understanding of the construction industry in terms of technology, education, and sustainability. The updated Pi-C is expected to assist decision makers in data-driven decision-making and strategy development in the construction industry.