CrowdSky: Skyline Computation with Crowdsourcing

In this paper, we propose a crowdsourcing-based approach to solving skyline queries with incomplete data. Our main idea is to leverage crowds to infer the pair-wise preferences between tuples when the values of tuples in some attributes are unknown. Specifically, our proposed solution considers three key factors used in existing crowd-enabled algorithms: (1) minimizing a monetary cost in identifying a crowdsourced skyline by using a dominating set, (2) reducing the number of rounds for latency by parallelizing the questions asked to crowds, and (3) improving the accuracy of a crowdsourced skyline by dynamically assigning the number of crowd workers per question. We evaluate our solution over both simulated and real crowdsourcing using the Amazon Mechanical Turk. Compared to a sort-based baseline method, our solution significantly minimizes the monetary cost, and reduces the number of rounds up to two orders of magnitude. In addition, our dynamic majority voting method shows higher a...

• Type: Journal
• Year: 2016
• Where: EDBT
• Keywords: Database, EDBT 2016