2020/10/25 by Hamza Salem, Salem, Hamza, Fabian Stephany +1
Social Sciences · Computer Science · #Social Media and Politics #Hate Speech and Cyberbullying Detection #Media Influence and Politics
paper · pdf · doi:10.48550/arxiv.2010.14627
Election prediction has long been an evergreen in political science\nliterature. Traditionally, such efforts included polling aggregates, economic\nindicators, partisan affiliation, and campaign effects to predict aggregate\nvoting outcomes. With increasing secondary usage of online-generated data in\nsocial science, researchers have begun to consult metadata from widely used\nweb-based platforms such as Facebook, Twitter, Google Trends and Wikipedia to\ncalibrate forecasting models. Web-based platforms offer the means for voters to\nretrieve detailed campaign-related information, and for researchers to study\nthe popularity of campaigns and public sentiment surrounding them. However,\npast contributions have often overlooked the interaction between conventional\nelection variables and information-seeking behaviour patterns. In this work, we\naim to unify traditional and novel methodology by considering how information\nretrieval differs between incumbent and challenger campaigns, as well as the\neffect of perceived candidate viability and media coverage on Wikipedia\npageviews predictive ability. In order to test our hypotheses, we use election\ndata from United States Congressional (Senate and House) elections between 2016\nand 2018. We demonstrate that Wikipedia data, as a proxy for\ninformation-seeking behaviour patterns, is particularly useful for predicting\nthe success of well-funded challengers who are relatively less covered in the\nmedia. In general, our findings underline the importance of a mixed-data\napproach to predictive analytics in computational social science.\n