Predictive Coding
You must have one of the following user roles to conduct these tasks:
- CORE Administrator
- All
- Case Manager
- Power User
Predictive Coding helps you find potentially relevant documents. It combines machine learning with human review: once reviewers have tagged documents relevant to a Predictive Coding-enabled review workflow, you add an iteration to the workflow to run Predictive Coding. Each time an iteration is added, the software makes a model out of the user-tagged input set and looks for other documents in the matter that have qualities similar to the model. The result is a list of workflow-specific documents suggested for review, each with a score corresponding to how closely the document compares to the model. The results are human-reviewed to confirm whether the suggested documents are relevant. Over the course of a review project, multiple Predictive Coding iterations are run to further refine the system’s ability to identify potentially responsive content.
Predictive Coding is best known as a process that statistically limits the volume of data subject to human review based on targeted levels of precision and recall. Estimation samples play a key role in this strategy as the system generates metrics from the reviewed sample and those metrics help the case team determine when to stop human review.
Predictive Coding can also be used informally as a strategic analysis strategy, generally known as More Like This Predictive Coding, where estimation samples may or may not be used.
This chapter discusses the following: