SUL AI Studio Experiments
Wednesday, January 23, 2019
11:00 am - 12:00 pm
East Asia Library, Room 224
In July of 2018 we launched the SUL AI Studio with the goal of surfacing projects within the library that might benefit from artificial intelligence techniques. In this session project teams from across the library will share the results of their experiments with AI.
Using Topic Modeling to Describe 19th Century Novels
Rebecca Wingfield, Arcadia Falcone, Javier de la Rosa, Scott Bailey
In 2011, Stanford acquired a collection of single-volume 19th century novels not widely held in major research libraries. The collection encompasses a wide range of genres (mystery, adventure, religious fiction, etc.) that were popular in the period. There is currently only minimal description of what each novel is about, making it difficult for researchers to understand what genres, themes, places, characters, and time periods are represented. This project explores to what extent topic modeling might be used to enhance subject access to the collection.
Automating Audio Transcription
Josh Schneider, DeBauche and Sarav Shah
This team has been working on a project to implement Google's Cloud Speech-to-Text tool to automate the transcription of audio recordings from the Library's Special Collections and University Archives including the Allen Ginsberg papers and KZSU Project South interviews. After working through some technical and logistical challenges with the tool itself as well as our workflow, we're excited to share our method for implementing the the tool and some of the transcriptions we have produced.
Lowering the bar to making CloudVision data usable by leveraging IIIF
Jessie Keck, Jack Reed, and Chris Beer
IIIF makes accessing images as easy as constructing a URL. Google CloudVision makes running images through AI based image analysis relatively easy as well. Our project aims to leverage the data returned from CloudVision in a way that will allow IIIF based viewers to be able to consume the image analysis data using existing open standards.
Image Recognition for Archaeological Research
Claudia Engel and Justine Issavi
This project focuses on the image repository of the Çatalhöyük Archaeological Project. After 25 years of research, the project has accumulated about 150,000 images. Extracting the wealth of relevant information from these images for research remains a challenge, as metadata are incomplete or inconsistent and researchers are often interested in information that is not captured at all. We have been exploring the use of AI for image tagging and archaeological object recognition (for example https://cengel.github.io/Catal-Vision-API/).
Similar Image Search: How Machine Learning Can Assist with Metadata Creation
Machine learning has allowed significant advances to be made with visual search giving metadata librarians new possibilities for automating and improving metadata creation workflows.