Research

My current research is the large scale analysis of news-media content. Media analysis has been the field of social scientists for a long time. Their studies typically focus on detecting specific biases over limited number of news-outlets and for small time periods, since most of the work is performed manually. Nowadays, the on-line presence of most mainstream media, the advances of modern machine learning and data mining techniques allows the automated analysis of vast numbers of outlets in large time frames. I work on detecting, explaining and understanding the patterns that can be found in the news media content.

Examples of my research findings and achievements:

Reconstruction of the news-outlets network, based on the sharing of content of news-outlets Relevant publications: [ECML/PKDD2010,WI-IAT2010]	An illustration of the global media network. Nodes are news-outlets that cover the same stories more than expected by chance. Outlets from the same country are coloured the same.
Prediction of popularity of news articles based on SVM ranking. Relevant publications: [AIAI2010]	Prediction accuracies of "popular" vs. "non-popular" articles for ten news outlets.
Comparing countries based on their media content. Relevant publications: [CIP2010]	Example of `Citations' network of EU countries for July 2009. Country A points to country B if B is mentioned `frequently' in news outlets of country A. .
Measuring biases in media content. Relevant publications: [WAPA2010]	Comparison of readability of different topics in large scale-across a vast amount of news articles, for hundreds of outlets and for over a year of continuous monitoring.
Learning of network topology. Relevant publications: [ECML/PKDD2010]	Three reference networks are constructed based on nodes' properties of the news-outlets network, i.e. language and country of origin, and media type. We show that they are able to predict network topology under a supervised learning scheme based on Generalised Linear Models.
Our group created a demo, namely Found in Translation, that presents some aspects of my work. My contributions were on text categorisation and building the demo pipeline. Relevant publication:[ECML/PKDD2010B].	We constantly monitor and compare the EU countries based on the topic biases that their media choose to cover.

- Multilabel classification of multilingual news articles [WI-IAT2010, ECML/PKDD2010B].
- Correlation of the network topology to other known networks