March 28, 2015 21:48 WSPC/INSTRUCTION FILE Tsapatsoulis˙Agathokleous˙final International Journal on Artificial Intelligence Tools c  World Scientific Publishing Company On the design of Social Voting Recommendation Applications Nicolas Tsapatsoulis, Marilena Agathokleous, Constantinos Djouvas Cyprus University of Technology, P.O. Box 50329, 3036, Lemesos, Cyprus, {nicolas.tsapatsoulis, costas.tziouvas}@cut.ac.cy mi.agathokleous@edu.cut.ac.cy Fernando Mendez Centre Democracy Studies Aarau, University of Zurich, 21 Kuttigerstrasse, Aarau 5000, Switzerland, fernando.mendez@zda.uzh.ch Received (Day Month Year) Revised (Day Month Year) Accepted (Day Month Year) Voting Advice Applications (VAAs) are online tools that match the policy preferences of voters with the policy positions of political parties or candidates. Designed to enhance the political competence of citizens, VAAs have become increasingly popular and insti- tutionally embedded in a growing number of European countries. While the traditional VAA relied on the stated position or academically coded position of parties/candidates, a recent innovation has been to introduce a social vote recommendation borrowing the basic principles of collaborative filtering. The latter takes advantage of the community of VAA users to provide a vote recommendation. This paper provides an overview of the social vote recommendation scheme and tackles three problems related to its optimal im- plementation in a real–world setting: (1) the number of samples required to train party models; (2) whether this number is affected by differences in characteristics between early users versus late users; and (3) whether generalizations can be derived across VAA applications in different countries. For our experiments we use three real VAA datasets based on elections in Greece 2012, Cyprus 2013 and Germany 2013. The corresponding datasets are made freely available to other researchers working in the areas of VAA and web based recommender systems. 1. Introduction Over the last decade so-called Voting Advice Applications (VAAs), which can be thought of as a type of vote recommendation system, have become a notable feature of the electoral landscape in many European countries 23 . The increasing popularity of VAAs can be gleaned from the sheer number of ‘recommendations’ produced in certain political settings –according to some analysts nearly 40 per cent of the electorate in the case of The Netherlands or 6.7 million ‘recommendations’ in Ger- many (see various country chapters in Cedroni and Garzia 2010 6 ). In view of this 1