Dynamic context extraction in personal communication applications

Data generated by video applications are rarely mined for context in order to augment and improve user experiences. In this paper, we propose an innovative approach to extract context from video streams dynamically. As a case study we apply this approach in a context based multimedia chat application. In particular, this paper discusses how we perform video stream analysis to recognize faces and logos; separate audio from video contents for further analysis; and mine text chat messages for keywords to infer contextual information at all levels. This allows us to recognize people, logos, as well as conversation topics and recommend videos on the fly. One key technique is that the chat application maintains different context spheres that can be selectively combined and intersected to provide a context sensitive and improved chat user experiences.