Construction safety knowledge sharing on YouTube from 2007 to 2021: Two-step flow theory and semantic analysis

•This study explores how YouTube facilitates construction safety knowledge sharing through the perspective of two-step flow theory.•Via social network analysis and semantic analysis, this study explored the video network and investigate how sentiment affects users’ viewing and commenting behaviours.•This study reveals that most top commenters were construction safety experts and staff from training institutes or companies, who attempted to market their brands via YouTube.•Harel-Koren Fast Multiscale layout algorithm results indicated that neither the opinion leaders nor the users had a strong will to interact with others.•Opinion leaders might use videos with negative emotions for effective knowledge sharing. This study explores how YouTube facilitates construction safety knowledge sharing. It contributes to knowledge by applying two-step flow theory, social network analysis, and semantic analysis to explore the video network and investigate how sentiment affects users’ viewing and commenting behaviours. By conducting data mining and interactive visualisation algorithms, this study analysed 500 construction safety videos from 2007 to 2021, 2000 users and 2608 comments about these videos. Most top commenters were construction safety experts and staff from training institutes or companies. They attempted to market their brands via YouTube. 3722-word pairs semantic analysis, indicated that most opinion leaders expressed their opinions positively. Half of the construction safety YouTube videos were uploaded in 2020 and 2021, indicating that YouTube social network and video making activities became prevalent during the COVID-19 pandemic. Besides, the US accounts for half of the top ten videos uploaders. Many YouTube videos mentioned social media such as “Facebook,” “Instagram”, “Twitter,” and “LinkedIn”, evidenced cross information sharing among social network platforms. Popular Indonesian words like “Bangunan” (building), “kerja” (work) and Ahmedabad (India’s industrial hub) evidenced the popularity of Indonesian and Indian in sharing construction safety knowledge via YouTube. Unlike many YouTubers in the entertainment sector who utilised it to interact with fans, Harel-Koren Fast Multiscale layout algorithm results of 277 YouTube video clusters revealed that neither the opinion leaders nor the users had a strong will to interact with others.