Abstract
The widespread usage of crowdsourcing systems, which leverage human intelligence to do computer-hard tasks, has created an urgent need to combine crowdsourcing with automated computation while optimizing performance issues, such as monetary cost. The Pareto-optimal query, which is a popular tool for multi-criteria decision-making services, is demanded by many crowdsourcing applications. Traditional researches require explicit attribute values where choices (or objects) are naturally totally ordered. However, it is hard for crowdsourcing workers to provide numerical values, and preferences are partially ordered in many real-life cases. In this paper, we study the problem of multiple Pareto-optimal queries over partial orders by leveraging crowdsourcing to obtain preference relations. The goal is to minimize the number of pairwise comparison questions, thereby reducing the monetary cost. We first introduce the concept of strict dominance where a strictly-dominated non-Pareto-optimal object can be safely pruned without false hits in the final results. We then present a preference tree-based query scheme, where objects in higher levels of this pre-constructed tree are more capable of (strictly-) dominating others. This tree structure enables the crowdsourcing platform to quickly identify non-Pareto-optimal objects and prune strictly-dominated non-Pareto-optimal objects. Extensive experiments demonstrate that the strictly-dominated non-Pareto-optimal objects take up more than 70% of the uniform data set, and the proposed approach effectively reduces the number of comparison questions.
Published Version
Talk to us
Join us for a 30 min session where you can share your feedback and ask us any queries you have
Schedule a call