Incremental Video Categorization Using Audio and Video Features

点击翻译按钮获取中文摘要

Incorporated by seasoned media executives Tessa Sproule and Katie MacGuire in 2014, Vubble is a media and education technology company that builds and offers innovative solutions for trustworthy digital video curation, assessment, and distribution. Vubble has a team of editors, journalists and teachers who curate quality online video content, selecting the best, most informative videos and assess their veracity using the Vubble Credibility Meter, which Vubble developed at MIT, in collaboration with the Nieman Journalism Lab, in February 2017. Vubble's editors add structured data to each video, through their 'data labelling' process-they tag content using their taxonomy, which includes categories such as 'topic', 'tone' and 'grade level'. Categorization and assignment of textual labels to the videos takes valuable editorial time and limits the number of videos an editor can process in a day. This manual process is a hindrance to Vubble's ability to scale its operations to meet the increasing needs of its current customers and grow its customer base. Vubble is partnering with Seneca to develop and implement an automatic system, customized for video curators, to categorize video material based on both visual and audio cues. Implementation of the system will enable Vubble to increase its productivity and address existing and new market demand, ultimately enabling the scaling of its business.