Algorithms for implicit delegation to predict preferences

Abstract

The lives of people are becoming more and more digitalized. With this their preferences over different topics like their favorite music or websites can be collected and analyzed. One possible goal of such an analysis is to create an overall ranking with the preferences of multiple people or countries. For this computational social choice comes into play with algorithms that take multiple rankings as input and output one ranking that represents the whole group as good and fair as possible, these algorithms are called social welfare functions. If some application wants then to create regular rankings that include the preferences of all its users there needs to be a way to predict the current preference rankings of users that did pause using the service. For this implicit delegation could be the solution as it takes previous preference data that is known and tries to create new rankings from them for the given user/person. For this thesis some algorithms were developed that try to accomplish this. One goal of these algorithms is to produce rankings that match the actual top-k preferences of a user or also called voter. As second goal it is attempted to replace multiple missing rankings and then use social welfare functions to aggregate them. Here it optimally produces a ranking that is as similar to an aggregated ranking with the real data as possible. For the similarity Kendall tau algorithms are used. To see how the implicit delegation methods work on real world data they are tested on real data sets from sources like Spotify.