Data enrichment using heterogeneous sources

A method for data enrichment, the method comprising:

• measuring, by an attribute relevance module that is executed by at least one hardware processor, relevance of an attribute to a data object to be enriched, the data object including the attribute including one of a known and an unknown value, wherein the relevance of the attribute includes a determination of a unique association of the attribute with the data object, and a determination of a discriminative property of the attribute with respect to instances of the data object based on evaluation of an entropy of a predetermined number of past values of the attribute;
• determining, by an output value confidence module that is executed by the at least one hardware processor, a confidence of a source based on utility of an output value of the source used for enrichment of the data object, the output value representing at least one of the known and unknown values of the attribute;
• determining, by the at least one hardware processor, assignment of the unknown value to the attribute by using the measured relevance of the attribute and the determined confidence of the source;
• and
• determining, by a source utility and adaptation module that is executed by the at least one hardware processor, utility of the source for enrichment of the data object by determining a confidence of input values provided to the source.

A non-transitory computer readable medium having stored thereon machine readable instructions to provide data enrichment, the machine readable instructions, when executed, cause at least one hardware processor to:

• measure, by an attribute relevance module, relevance of an attribute to a data object to be enriched, the data object including the attribute including one of a known and an unknown value;
• select, by a source selection module, a source for enrichment of the data object from a plurality of sources by at least one of
• determining whether the source includes a fitness score that exceeds a predetermined threshold, and determining the fitness score based on input attributes to the source, output attributes from the source, and a number of times the source is selected for enrichment of the data object,
• accounting for an uncertainty in an input to the source and an uncertainty in an output from the source, and
• using a half-life factor to downgrade usefulness of the source based on a number of times the source is called;
• determine, by an output value confidence module, a confidence of the source based on utility of an output value of the source used for enrichment of the data object, the output value representing at least one of the known and unknown values of the attribute;
• determine assignment of the unknown value to the attribute by using the measured relevance of the attribute and the determined confidence of the source;
• and
• resolve, by an ambiguity resolution module, ambiguities in a plurality of output values that are generated by the source used for enrichment of the data object based on generation of a plurality of enriched data objects based on each of the plurality of output values that are generated by the source.