Table 1.

Review of current approaches for each data pipeline stage.

Current solutions			Domain expert role
Data curation
	Data integration
		Schema matching [138-143] Interactive integration [144,145] Webtables integration [146-151] Machine learning [46-49]		Domain experts are needed to validate results of integration, and interactively correct automated methods, which can then update their algorithm
	Data discovery
		Attribute search [58,59,152,153] Interactive querying [55,62-64] Visual analytics [56,57]		Domain expert feedback is needed to finalize the analysis data set
Data cleaning
	Error fixes
		Rule-based [69-77,154] Crowdbased [18,79,155,156] Knowledgebase [80-82] Machine learning [83-85] Functional dependency [15-23,25-54,58-165]		Domain expert input can be used to identify and fix errors
	Augmentation
		Machine learning [97,166,167] Interactive [28,98,168-170]		Domain experts can augment missing data with domain-specific rules
	Transformation
		Programming by example [102,103] Interactive rules [4,99-101] Foreign-key detection [153,171-175]		Domain experts can restructure the data to make it semantically valid
Data analysis
	Exploration
		Optimize performance [176-179] Optimize insight [126,180,181] Provenance [182,183] Visualizations [5,7,116-118,184-188]		Domain experts interact with summaries and outliers to draw insight
	Explainable
		Systems [189,190] Visualizations [9,12,29,136] Empirical studies [10,11,13,14]		Domain experts inform the model design to ensure explainability