Visual exploration of topics in multimedia news corpora

Abstract

As news contents grow daily, the demand for tools to help users make sense of large document corpus will continuously be on the increase. Such tools will particularly be useful for journalist and ordinary users who intend to explore large collection of news documents for various analytical tasks. When users attempt to explore documents, they are usually in search for a particular topic of interest, or to compare various topics for similarity, or to see when in time a particular topic was discussed or to explore the distribution of a topic over time or to see how frequent a particular topic was discussed in the corpus or in general to test a particular hypothesis. Existing tools fall short in providing effective and suitable interaction mechanism to enable users answer these questions in a single application framework. In this paper we presented a framework that gives users the opportunity to easily answer questions relating to their exploratory tasks. We developed new visual elements and augment them with existing interfaces to provide users with ample options and flexibility to explore multimedia news corpus from different angles depending on their analytic tasks. Our method uses machine learning for topic extraction, clustering and word cloud generation. Our approach effectively combines both overview + detail and focus + context schemes to enrich users experience with exploring large collection of multimedia news documents. Our framework ensures synchronization of the various visual interfaces to provide immediate feedback on user's interactions. To demonstrate the effectiveness of our approach, we presented some realistic use cases from the perspective of a news analyst. And based on our observations, we identified some possible directions for future studies.