Review and Training Predict

Review = Training

What do we mean by review equals training? Reviewing means that a human reviewer is making a decision and coding the decision field of a document based on that decision. Training means that throughout this review process, we teach Predict what is positive (in our scenario, this means Responsive) and what is negative (Not Responsive) based on those human decisions. Then, it will examine the contents of the documents, find others similar to them, and rank the non-reviewed documents based on those human decisions. Training relies on review, and therefore all review is training.

All TAR systems need people to make decisions. With Insight, any reviewed document, meaning any document reviewed by any method in Insight, will be submitted to Predict to help rank the rest of the population. There is no need to select a random set to get started.

The only thing Predict verifies is whether:

• The document is in the Predict collection, and

• The decision field has a coded value

Many cases start with documents that have already been reviewed. If these documents are part of the collection you have already created, they will automatically be sent to Predict to help rank the non-reviewed documents. This happens when you add them to the collection and Create the Graph database.

If no documents in the collection have been reviewed (contain coding on the decision field), you can send them to a reviewer by any review method you choose (through Dynamic Folders, Static Folders, the Review Module, or by running a search).

All review in Insight Predict trains the Predict engine how to rank documents within a collection.