Network-centric Approaches to the Exploration of News Streams

The interconnectedness and frequency of published news articles is both a blessing and a curse for the contemporary news consumer. On the one hand, it is easy to get swept away by the deluge of articles in the news stream and miss the big picture. On the other hand, given the right amount of automated preprocessing, it is also much easier to discover connections that would otherwise be hidden to the unaided reader. In this talk, I discuss two network-centric approaches that provide a novel perspective on the interconnectedness of news articles. In the first half, I introduce the concept of implicit entity networks and show how they can be obtained from large streams or collections of unstructured texts. Focusing on the entity types of location, organization, person, and date, I highlight applications for such networks in the exploration and visualization of events and topics in entangled news streams. Since implicit entity networks are highly dependent on named entity extraction and linking, I subsequently discuss their benefits and drawbacks in relation to word embeddings. In the second half, I adopt a coarser view on network structures in news by focusing on news citation networks as a novel perspective on information propagation in news. After a brief introduction into the concept of news citation networks, I show how their structure simultaneously ties them to scientific citation networks and sets them apart. In addition to some preliminary results in document dating based on the network structure of news, I discuss potential implications for the detection of ideological bubbles in news publishing.

Andreas studied computer science with a minor in computational linguistics at Heidelberg University, where he received his Master of Science degree in December 2014. Currently, he is working as a research assistant and PhD student in the Database Systems Research group at Heidelberg University under the supervision of Michael Gertz. His interests broadly cover the intersection of information retrieval, natural language processing, and complex network analysis. His recent projects have focused on the extraction and analysis of network representations of annotated textual sources to bridge the gap between fully structured knowledge graphs and entirely unstructured texts.