Training of Predictive Coding Systems Fosters Debate

http://ow.ly/9U5eC

An article by Evan Koblentz posted on law.com on the LTN webpage.

This article discussed predictive coding technology, and other similar technology assisted review processes.

The article states, "At issue is the method for measuring accuracy of the software's output. But a secondary debate may have wider long-term impact -- what's the best method for training predictive coding systems in the first place?

Training is the process of teaching the software, for every document review, which pieces of electronically stored information are responsive and which are not. It initially requires human input to determine relevancy, but at a certain point the computer takes over. A predictive coding system's final results can be more accurate than human reviewers, its advocates claim."

The article then examines some of the workflow processes followed by various service providers, and the methods used to provide more accurate results.  The article points out some of the limitations in the process, and the need for accurate input from the human coders that are setting up the process at the outset of a specific case.

P.S.  What is not referenced in the article are two important metrics that can be used to measure the effectiveness of any review:  Precision and Recall. In order for a review to be truly effective, the precision and recall rates must both be factors to consider.

Precision: the extent to which only responsive documents are captured during an attorney review.

Recall: the extent to which all responsive materials are captured in an attorney review.