Mining Multimedia Data for Meaning

Abstract

The explosion of images, video and multimedia is creating a valuable source for insights. It can tell us about things happening in the world, give clues about a persons preferences or experiences, indicate places of interest in a new town, and even capture a rolling log of our history. But, as a non-traditional source for data mining, there are numerous challenges to be overcome in order to handle the volume, velocity and variety of multimedia data in practice. In this talk, we review several application areas across Web, social media, mobile and safety/security and show how they benefit from mining of multimedia data. We review novel approaches for modeling semantics and automatically classifying visual contents and demonstrate examples in the context of IBM Multimedia Analysis and Retrieval System (IMARS).