Artificial Intelligence and Digitalisation in the Construction Industry: A Scientometric Analysis of Structural Integration, Thematic Acceleration and the Sustainability Integration Gap
Description
Purpose – This study maps the intellectual structure of research on artificial intelligence (AI) and digitalisation in the construction industry, moving beyond descriptive science mapping to a network-analytic diagnosis of how the field is structurally integrated, how rapidly its themes are accelerating, and whether its sustainability dimension is structurally embedded in the field. Design/methodology/approach – A corpus of 613 Scopus-indexed documents (2016–2026) was analysed through bibliometric performance analysis, keyword co-occurrence mapping in VOSviewer, and a reconstructed co-occurrence network subjected to multi-measure centrality analysis, cluster-cohesion estimation, modularity computation, and temporal emergence detection. Findings – The field has grown at a compound annual rate of 39.6 per cent to a corpus h-index of 73. Betweenness-centrality analysis reveals a hub-and-spoke architecture in which construction industry is the singular structural bridge (betweenness 0.98) despite high network density (0.77). Cluster-cohesion analysis uncovers a descending integration gradient across the field's four thematic clusters, from a well-integrated digital-BIM core (cohesion 0.561) to a barely-integrated sustainability cluster (0.106). Emergence analysis shows that machine-learning and deep-learning methods are accelerating most rapidly, while sustainability themes are both the least cohesive and among the slower-accelerating. The study terms this divergence the sustainability integration gap. Originality/value – This is the first scientometric study of AI and digitalisation in construction to combine multi-measure centrality, cluster-cohesion estimation, and emergence detection, and the first to demonstrate empirically that the field is deepening technologically while leaving its sustainability dimension structurally detached.
Files
Steps to reproduce
Research Methodology The study employed a scientometric review design integrating bibliometric performance analysis with network-analytic science mapping (Donthu et al., 2021). Performance analysis quantified the productivity and impact of the field's documents, authors, sources, and countries, while science mapping reconstructed and analysed the keyword co-occurrence network through formal network metrics. Scopus was selected as the data source for its comprehensive coverage of engineering literature and its suitability for bibliometric analysis (Pranckute, 2021). The search combined the constructs of artificial intelligence or digital technology with construction industry in the title, abstract, and keyword fields, restricted to 2016–2026 accumulating an initial search of 2,306, to the Engineering and Environmental Science subject areas (1,851 documents), to journal articles and conference papers, and to English-language documents (984 documents). After successive screening, the final corpus comprised 613 documents, of which 373 were journal articles and 240 were conference papers. Minimum thresholds were applied: 100 citations per document (42 of 613 qualifying), four documents per author (19 of 1,904 qualifying), eight documents per source (15 of 265 qualifying), 20 documents per country (11 of 74 qualifying), and 30 occurrences per keyword (28 of 4,738 qualifying). These thresholds follow established bibliometric practice and yield analytically tractable networks while retaining the field's substantive structure. The keyword co-occurrence network was reconstructed from the raw bibliographic records, with nodes representing the 28 qualifying keywords and edges weighted by co-occurrence frequency. The reconstruction was validated against the total link strengths reported by VOSviewer. Four centrality measures were computed: weighted degree, capturing cumulative co-occurrence strength; degree centrality, capturing the breadth of co-occurrence; betweenness centrality, computed on inverse-weight distances to identify structural bridges; and eigenvector centrality, capturing connection to well-connected nodes. At the network level, density and modularity were computed for the four-cluster partition identified through co-occurrence clustering. The search query was TITLE-ABS-KEY ( "artificial intelligence" OR "Digital Technology" AND "Construction Industry" ) AND PUBYEAR > 2015 AND PUBYEAR < 2025 AND ( LIMIT-TO ( SRCTYPE , "j" ) OR LIMIT-TO ( SRCTYPE , "p" ) ) AND ( LIMIT-TO ( EXACTKEYWORD , "Construction Industry" ) OR LIMIT-TO ( EXACTKEYWORD , "Artificial Intelligence" ) ) AND ( LIMIT-TO ( DOCTYPE , "ar" ) OR LIMIT-TO ( DOCTYPE , "cp" ) ) AND ( LIMIT-TO ( SUBJAREA , "ENGI" ) OR LIMIT-TO ( SUBJAREA , "ENVI" ) ) AND ( LIMIT-TO ( LANGUAGE , "English" ) )
Institutions
- Durban University of TechnologyKwaZulu-Natal, Durban