Title: Forecasting election results by studying brand importance in online news

Abstract: This study uses the semantic brand score, a novel measure of brand importance in big textual data, to forecast elections based on online news. About 35,000 online news articles were transformed into networks of co-occurring words and analyzed by combining methods and tools from social network analysis and text mining. Forecasts made for four voting events in Italy provided consistent results across different voting systems: a general election, a referendum, and a municipal election in two rounds. This work contributes to the research on electoral forecasting by focusing on predictions based on online big data; it offers new perspectives regarding the textual analysis of online news through a methodology which is relatively fast and easy to apply. This study also suggests the existence of a link between the brand importance of political candidates and parties and electoral results.