A Rating-Ranking Method for Crowdsourced Top-k Computation

摘要: Crowdsourced top- k computation aims to utilize the human ability identify Top- objects from a given set of objects. Most existing studies employ pairwise comparison based method, which first asks workers compare each pair and then infers results on results. Obviously, it is quadratic every object these methods involve huge monetary cost, especially for large datasets. To address this problem, we propose rating-ranking-based approach, contains two types questions ask crowd. The rating question, crowd give score an object. second ranking rank several (e.g., 3) Rating are coarse grained can roughly get object, be used prune whose scores much smaller than those Ranking fine refine scores. We unified model questions, seamlessly combine them together compute also study how judiciously select appropriate or assign coming worker. Experimental real datasets show that our method significantly outperforms approaches.