Speculative Recommendation: Reframing AI for Interpretive Practice in the Digital Humanities

Abstract

This project investigates the potential of recommender systems as a generative tool in humanities research using the case of aesthetic-driven recommender system focused on animated films. Leveraging still frames from trailers and a fine- tuned convolutional neural network (VGG16), we extracted aesthetic features to develop a recommender system that uses visual characteristics as the sole input. The system demonstrates how recommender systems can support the investigation of high level concepts such as influence, identifying both expected and surprising latent connections between films on the basis of visual similarities. This approach has broader implications within DH, showing how recommender systems can facilitate exploratory, interpretive research, how the stochasticity of AI models is well suited to support the interpretative aspect of humanistic scholarship, and exploring the potential of interdisciplinary cross-pollination in DH research.