Diverse user selection for opinion procurement

Many applications maintain a repository of user profiles with semantically rich information on each user. Such repositories have a potential of allowing active opinion procurement: reaching out to users to ask for their opinions on different topics. An important desideratum of the procurement process is that it targets a diverse set of users. To realize this potential, we present Podium: a first framework, to our knowledge, that supports the selection of diverse representatives in presence of high-dimensional, semantically rich user profiles. We demonstrate that data dimensionality is a challenge for both defining and achieving diversification. We address these challenges by proposing a lightweight, flexible notion of diversity that in turn allows explanations and customization of diversification results. We show that the problem of finding an optimally diverse user subset is intractable, and provide a greedy algorithm that computes an approximate solution. We have implemented our solution in a system prototype and tested it on real-world crowdsourcing platform data. Our experimental results show that Podium is effective in selecting users with diverse properties, and in turn that the opinions of these users are diverse according to multiple metrics.