Guided data repair

By Mohamed Yakout, Ahmed K. Elmagarmid, Jennifer Neville, Mourad Ouzzani, Ihab F. Ilyas. 2011

233 citations recorded by the source corpus.

In this paper we present GDR, a Guided Data Repair framework that incorporates user feedback in the cleaning process to enhance and accelerate existing automatic repair techniques while minimizing user involvement. GDR consults the user on the updates that are most likely to be beneficial in improving data quality. GDR also uses machine learning methods to identify and apply the correct updates directly to the database without the actual involvement of the user on these specific updates. To rank potential updates for consultation by the user, we first group these repairs and quantify the utility of each group using the decision-theory concept of value of information (VOI). We then apply active learning to order updates within a group based on their ability to improve the learned model. User feedback is used to repair the database and to adaptively refine the training set for the model. We empirically evaluate GDR on a real-world dataset and show significant improvement in data quality using our user guided repairing process. We also, assess the trade-off between the user efforts and the resulting data quality.

Subfields: Artificial Intelligence, Computer Science Applications, Management Science and Operations Research

How researchers see this paper

The researchers whose trusted networks rank it highest, strongest advocate first. Every researcher gets a different order.

In Artificial Intelligence

In Computer Science Applications

In Management Science and Operations Research

Citations in the corpus

AlphaClean: Automatic Generation of Data Cleaning Pipelines (2019, methodology): “Machine Learning in Data Cleaning: Machine learning has been widely used to improve the efficiency and/or reliability of data cleaning [19, 48, 49].”
BoostClean: Automated Error Detection and Repair for Machine Learning (2017, background): “There are a number of other works that use machine learning to improve the efﬁciency and/or reliability of data cleaning [21,51,52].”
ActiveClean: An Interactive Data Cleaning Framework For Modern Machine Learning (2016, methodology): “We will first describe ActiveClean and overview the entire framework, and a detailed description of the research challenges and algorithms can be found in [8].”
HoloClean: Holistic Data Repairs with Probabilistic Inference (2017, background): “ribute A n 2 A . We assume that errors in D occur due to inaccurate cell assignments,andweseektorepairerrorsbyupdatingthevaluesofcellsin D . This is a typical assumption in many data cleaning systems [12, 13, 31, 40]. For each cell we denote by vc its unknown true value and by vc its initial observe”
BigDansing: A System for Big Data Cleansing (2015, background): “There are also several efforts to include users (experts or crowd) in the cleaning process [33,40].”
Holistic data cleaning: Putting violations into context (2013, background): “On the other hand, in academia, researchers are investigating declarative, constraintbased rules [4], [5], [13], [8], [11], [12], [21], [18], which allow users to detect and repair complicated patterns in the”
Data Lake Management: Challenges and Opportunities (2019, background): “Logical and relational data cleaning typically requires correct schema information including integrity constraints [34].”
ActiveClean: Interactive Data Cleaning While Learning Convex Loss Models (2016, methodology): “There are a number of other works that use machine learning to improve the efﬁciency and/or reliability of data cleaning [38, 37, 16].”
Big data challenge: a data management perspective (2013, background): “In [14, 16], several algorithms are proposed to automatically incorporate user feedbacks and update existing data integration programs.”
Trends in Cleaning Relational Data: Consistency and Deduplication (2015, methodology): “Guided Data Repair Guided data repair (GDR) [126] incorporates user feedback in the data cleaning process to enhance and accelerate automatic data repairing techniques for CFDs while minimizing user involvement.”
NADEEF: a commodity data cleaning system (2013, methodology): “Deployment and extensibility: Although many algorithms and techniques have been proposed for data cleaning [5, 14, 29], it is difficult to download one of them and run it on the data at hand without tedious customization.”
Towards reliable interactive data cleaning: a user survey and recommendations (2016, background): “While human-in-the-loop data cleaning systems have been extensively studied [10] [12] [17] [19] [33] [37] [41], a key missing piece is a detailed study of how data analysts interact with such systems.”
Cleaning Crowdsourced Labels Using Oracles For Statistical Classification (2018, background): “One category is to leverage humans (either crowd workers or experts) to solve a particular data-cleaning problem, such as entity resolution [9,14,39,41], missing value imputation [30], and data repairing [8,42].”
Towards Automated Data Cleaning Workflows (2019, methodology): “GDR uses active learning to choose the correct update suggested by user-deﬁned functional dependencies [33].”
KATARA: A Data Cleaning System Powered by Knowledge Bases and Crowdsourcing (2015, methodology): “To improve the accuracy of data repairing, experts have been involved as first-class citizen of data cleaning systems [19, 35, 44], high quality reference data has been leveraged [19, 24, 42], and confidence values have been placed by the users [18].”
ActiveClean: Interactive Data Cleaning For Statistical Modeling (2016, background): “Comparison to Active Learning: The motivation of ActiveClean similar to that of Active Learning [16,36] as both seek to reduce the number of queries to an analyst or a crowd.”
A sample-and-clean framework for fast and accurate query processing on dirty data (2014, background): “Particularly, in order to obtain reliable cleaning results, many datacleaning techniques need humans to get involved [24,36,58].”
Don't be SCAREd: use SCalable Automatic REpairing with maximal likelihood and bounded changes (2013, unknown): “, [22]) to inspect very small number of the least beneficial updates after each iteration.”