Legal teams preparing for predictive coding or Technology Assisted Review (TAR) seek solutions to increase the likelihood of accuracy in the artificially intelligent coding. The resulting accuracy depends upon the accuracy and consistency of the review of sample documents used by the system for the automatic document profiling. Subsequent sample reviews may be required if the initial review does not produce good preliminary test results, thereby increasing the number of documents to review and the time to completion of the final coding process.
Capital Novus provides technology in eZSuite’s eZReview to mitigate inconsistency in coding and to enable CN’s ADP TAR Predictive Coding technology to generate the best results possible. Like the two players who beat a computer at chess in 2005 by combining human intellect with computer processing, eZReview provides tools to allow the human reviewer to do the parts that they do best and the computer to do the parts that technology does best.
eZReview provides text file size searching that can identify files with no text to eliminate documents that do not have the text required for the analysis of the predictive coding algorithms. These documents can be pictures, images, or other media files. They can be system files or auto-generated emails that are sent out to the mass public. The removal of these documents from sample sets eliminates the opportunity for skewed results based on documents that are not conducive to analysis.
This technology works best with a small team of expert reviewers assigned to the review to minimize disparate calls on similar documents and increase the likelihood of consistent coding. Ensuring that the same 2-3 experts review the initial control sample and the training samples yields best results in this environment.
System algorithms require the team to review based purely on document content. The document body and key metadata values function as the review sources. Factors such as family members, email thread leafs, or related documentation should not be considered when making the sample coding decisions.
Following the review of the control and training sample sets, the final system checkpoint includes eZReview’s ADP Conflict Decision Reporting. Dynamic reports provide comparisons and hyperlinks to address conflicting coding decisions between similar documents. Reports provide conflicting coding designation identification for exact and near duplicates.