CrowdAidRepair: A Crowd-Aided Interactive Data Repairing Method

Abstract

Abstract. Data repairing aims at discovering and correcting erroneous data in databases. Traditional methods relying on predefined quality rules to detect the conflict between data may fail to choose the right way to fix the detected conflict. Recent e↵orts turn to use the power of crowd in data repairing, but the crowd power has its own drawbacks such as high human intervention cost and inevitable low e ciency. In this paper, we propose a crowd-aided interactive data repairing method which takes the advantages of both rule-based method and crowd-based method. Particularly, we investigate the interaction between crowd-based repairing and rule-based repairing, and show that by doing crowd-based repairing to a small portion of values, we can greatly improve the repairing quality of the rule-based repairing method. Although we prove that the optimal interaction scheme using the least number of values for crowd-based repairing to maximize the imputation recall is not feasible to be achieved, still, our proposed solution identifies an e cient scheme through investigating the inconsistencies and the dependencies between values in the repairing process. Our empirical study on three data collections demonstrates the high repairing quality of CrowdAidRepair, as well as the e ciency of the generated interaction scheme over baselines.

Web-ADARE: A web-aided data repairing system

“…The NADEEF system [8] allows the users to specify data quality rules and how to repair erroneous data through writing code that imple-25 ments predefined classes. In our published conference paper [9], we develop Crow-dAidRepair, a novel Crowd-Aided Data Re-pairing approach, which performs crowdbased and rule-based repairing alternatively for achieving a high repairing quality at the minimum crowd cost. On the one hand, crowdsourcing can significantly improve the quality of the data, but on the other hand, although some efforts are made, it still…”