Συστήματα συστάσεων σε ηλεκτρονικούς συμβούλους ψήφου

Abstract

Recommender Systems (RSs) are software tools and techniques that make suggestions for products and services to potentially interested users. Voting Advice Applications (VAAs) are online tools that compare the political views of users with the political positions of political parties or candidates and indirectly suggest to the user the candidate or party with similar views. This thesis uses techniques and methods from RSs to the EU-wide VAA “EUvox” 2 to explore specific research questions. At the beginning, the subject, purpose and objectives of the thesis are introduced. Then, reference is made to the existing literature on RSs and VAAs. The rest of the work presents the methodology used and the experiments conducted in order to answer the research questions. The results of the experiments are compared to the existing research. The results showed, among other things, that non-linear methods perform better than other machine learning techniques, and the use of supplementary questions increases the performance of the voting prediction. Matrix factorization techniques can be implemented with great success in estimating supplementary questions that were not answered by users. There is heterogeneity within the parties, and many users give similar responses with other users who support different parties. As far as the identity of the European voter is concerned, he seems to be a man, young, highly educated and very interested in politics. Male educated users tend to be more interested in politics than women or those with a low level of education. As well as the age group does not seem to be an important factor for the user’s interest in politics.