| BRIDGE addresses access to television archives. Television is not an isolated medium, but a key node in a complex information landscape. From an exclusive focus on producing radio, television and publishing program guides, broadcasting organizations have moved to cross-media production companies where users are directly involved in online follow-up discussions. To be meaningful, these additional materials need to be cross-linked to the traditionally archived television materials so as to arrive at a meaningful web of information. The project will develop, test and deploy efficient, effective and robust methods for automatically creating meaningful links between dynamically expanding archived television data and other information sources that are centered around entities, themes and events. Our work takes the identification of entities, themes and events as its starting point. For entities we will adopt a data-driven approach, which will utilize the Sound and Vision cataloging system and Wikipedia. For theme classification, we aim to develop scalable methods to automatically acquire training data on a broad spectrum of topics. For event detection and normalization we will mainly use the repetition of stories covering the same event by leveraging their intrinsic threaded structure. Then, we will use generative and discriminative approaches to create and classify links using entities, themes, and events as link sources, while link targets can be documents or arbitrary text snippets. A historic study into the position of migrants using the developed tools will complement and inform our system-centered work. |
