Electoral Campaigns and Relation Mining: Extracting Semantic Network Data from Newspaper Articles

Abstract

ABSTRACT Among the many applications in social science for the entry and management of data, there are only a few software packages that apply natural language processing to identify semantic concepts such as issue categories or political statements by actors. Although these procedures usually allow efficient data collection, most have difficulty in achieving sufficient accuracy because of the high complexity and mutual relationships of the variables used in the social sciences. To address these flaws, we suggest a (semi-) automatic annotation approach that implements an innovative coding method (Core Sentence Analysis) by computational linguistic techniques (mainly entity recognition, concept identification, and dependency parsing). Although such computational linguistic tools have been readily available for quite a long time, social scientists have made astonishingly little use of them. The principal aim of this article is to gather data on party-issue relationships from newspaper articles. In the first stage, we try to recognize relations between parties and issues with a fully automated system. This recognition is extensively tested against manually annotated data of the coverage in the boulevard newspaper Blick of the Swiss national parliamentary elections of 2003 and 2007. In the second stage, we discuss possibilities for extending our approach, such as by enriching these relations with directional measures indicating their polarity.