Construct relation extraction from structural equation modeling papers

Knowledge repositories can facilitate the process of identifying relevant previous work. However, populating such repositories is a time-consuming and tedious task. 'DISKNET', a repository focused on information systems papers, requires all construct definitions and relations contained in an article. Both, the text and images of the articles can be leveraged to extract construct relations through large language models and computer vision models, where each approach poses its own challenges. Current results suggest that knowledge extraction could be augmented to reduce the amount of human effort necessary.