Identifying Differences in News Coverage Between.pdfVIP

Identifying Differences in News Coverage Between Cultural/Ethnic Groups CHARLES B. WARD, MIKHAIL BAUTIN, and STEVEN SKIENA1 Department of Computer Science Stony Brook University Stony Brook, NY 11794-4400 1. INTRODUCTION Interactions between distinct ethnic/cultural groups comprise one of the dominant social forces shaping our world. Accurately quantifying the nature of these interactions provides essential data for social science research, in fields as diverse as history, political science, and international relations. In this paper, we report on an effort to use computational analysis of long-scale text streams (newspapers) to measure differences in the reference frequency and sentiment associated with distinct ethnic/cultural groups. Our methodol- ogy is as follows. Using two distinct and orthogonal classification methods (ethnic name analysis and co-reference association), we identify likely ethnicities and nationalities for each person mentioned in the news. By aggregating reference/sentiment counts across all members of a cultural/ethnic group, we obtain strong-enough signals to detect and measure interesting trends. In particular, we report on the following contributions: —Methods for Ethnicity and Nationality Detection – We develop a method to associate primary nationalities for distinct named entities (people) based on a statistical analysis of geographic collocations. We also briefly review our methods for name ethnicity detection based on surname / given name frequencies and k-mer analysis, which are more thoroughly presented in [Ambekar et al. 2009]. Here we validate both methodologies, and demonstrate how they can work together to measure ethnic divisions in regional and national contexts. —Geographic News Analysis of Cultural Groups – We demonstrate that our methods coupling ethnicity/nationality detection with large-scale news analysis produce striking