Automatic Document Profiling (ADP) 2.0 has been introduced in eZReview which supports “Continuous Active Learning (CAL)”. The ADP 2.0 introduces continuous ranking and continuous learning that brings greater amount of savings in review cost and time and meantime making the review process smooth and streamlined.
In CAL, the judgments made by the reviewer while reviewing the document set are continuously fed back to the system from which it can ‘learn’ and continuously rank the remaining documents based on the judgments. When the reviewer gets the next batch, the documents based on latest continuous ranking are presented. Thus reviewers get efficiently ranked documents every time that improve the review process and result in higher recall rates.
Two new options like Top Ranked and Uncertain introduced in ADP 2.0 will help in regular feeding by machine of highest ranked documents by primary tag or of documents about which the model is not certain to make a definitive ranking respectively while creating seed and training set.
Continuous Active learning in ADP 2.0 will help the machine predict and classify the documents accurately. Continuous Active learning will help in informed classification of documents where there is uncertainty as machine is able to learn from huge pool of examples in the responsiveness spectrum. The machine will be able to better predict how to exactly differentiate between responsive and nonresponsive documents.