An Attempt at Metadata Enhancement through Machine Learning
Peters, Todd C.
This presentation will share what learned about machine learning and applicability to generate metadata to enhance discoverability during a pilot project. Object detection through neural networks is a rapidly developing field. Using machine learning large sets of images can be analyzed, objects detected and classified. We used the pretrained models COCO, Inception, ResNet, VGG19, and Xception to classify objects in images in our San Marcos Daily Record newspaper negative collection. Our initial use of these models did not yield usable metadata, however it did provide a useful first step into machine learning and knowledge to develop future research.
machine learning, metadata, discovery
Peters, T., & Long, J. (2022). An attempt at metadata enhancement through machine learning. Presented at the Texas Conference on Digital Libraries, Austin, Texas.