Data extraction confidence attribute with transformations

摘要

A data extraction system for receiving and scanning documents to generate ordered input for storage in a database employs a non-linear statistical model for a data extraction sequence having a plurality of transformations. Each transformation transitions an extracted data value in various forms from a raw data image to a computed data value. For each transformation, a confidence model learns a confidence component for the particular transformation. The learned confidence components, generated from a control set of documents having known values, are employed in a production mode with actual raw data. The confidence component corresponds to a likelihood of transformation accuracy, and the confidence model aggregates the confidence components to compute a confidence for the extracted data value. A database stores the extracted data value labeled with the computed confidence attribute for subsequent use by an application employing the extracted data.