Entry Date:
July 15, 2014

Glue

Principal Investigator Andrew Lippman


Glue is a prototyping engine to support news and narrative analysis. The system works by coordinating the flow of media among an extensible set of asynchronous python processing modules. The growing set of existing modules analyzes web pages, video, and exogenous data such as tweets and creates fine-grained metadata, including frame-by-frame analysis for video. We use this to organize material for presentation, analysis, summarization. Currently, the system provides named-entity extraction, audio expression markers, face detectors, scene/edit point locators, excitement trackers, and thumbnail summarization. Glue includes a video recorder and processes 14 DirecTV feeds as well as video content crawled from the web. Video is retained dependent on storage capacity and the database is permanent. Glue is the metadata driver for most Ultimate Media projects -- a “digestion system” for mass media.