Cities are increasingly reusing industrial heritage as part of cultural and creative regeneration strategies. However, designers and decision-makers face the challenge of determining which features and elements of industrial heritage are more perceived and preferred by the general public and will shape the future character of the site. The availability of large numbers of user-generated photos (UGPs) on social media provides a new human-centred and grassroots lens for the above issue. This article proposes to analyze the content of UGPs to measure common regularities in people's perceptions of industrial heritage sites. The article focuses on three cases of event-driven industrial heritage transformation: Shougang Steel Factories in Beijing as the Big Air venue of the Beijing 2022 Olympic Winter Games, Parco Dora in Turin as the site of the Kappa FuturFestival since 2012, and Cockatoo Island in Sydney as the site of the Sydney Biennale since 2008. The analysis is conducted in three steps: collecting UGPs from social media sites with a time range from pre-event to post-event to compare the transition of perceived images; clustering the photos based on visual similarity using deep learning models; and extracting commonly perceived image features, including landscape scenes, specific buildings/facilities, and certain elements.