Social media data and computer vision in social impact assessment: Understanding human dimensions and cultural ecosystem services in hydroelectric landscapes
Abstract
Social media data has proven to be a valuable resource for assessing social impacts, especially when combined with rapid advancements in artificial intelligence technologies. This combination enables comprehensive analyses of larger datasets than traditional methods allow. This dissertation assessed the social impacts of three hydroelectric dams and their reservoirs by employing two computer vision models to analyze large social media imagery datasets. This study explored two new methods: 1) it utilized a ready-to-use pre-trained computer vision model and a natural language processing model to assess landscape images sourced from Instagram and geo-tagged to study areas; and 2) it explored the feasibility of training a computer vision model to classify images based on CES coding themes. The results showed compelling patterns of landscape uses and values. The findings also suggested that computer vision technologies can significantly enhance the social impact assessment toolkit, revealing meaningful patterns and implications for projects in practice.