TAR For Smart People Outline - Chapters 13 and 14

TAR For Smart People Outline - Chapters 13 and 14

March 25, 2017

Here's the last installment in my outline of John Tredennick's 'TAR for Smart People'.   I last posted an installment on February 10, 2017.   This night's installment is on Chapter 13, 'Case Study: Using Insight Predict for Review of Rolling Opposing Party Productions' and Chapter 14, 'Case Study: TAR Does Double Duty in a Government Probe'. 

 

13. Case Study: Using Insight Predict for Review of Rolling Opposing Party Productions Insight Predict Finds 75% of Hot Docs While Cutting Review 92%

 

A. Challenge: Find Hot Documents in Opponent’s Rolling Productions - Prioritized Review with Insight Predict

   1. Insight Predict - Uses Continuous Active Learning to rank each new document as it is added.  

   2. Insight Predict can handle document productions with a low richness. 

   3. New ranking rounds were run every 10 minutes.   Richness increased from 1% to 7%.  70% of hot documents found after reviewing only 8% of production.  

 

B. Contextual Diversity vs. 'Hide the Ball'

    1. Samples the production to insure there are no pockets of documents that escape human review.  A pocket of new documents unlike anything reviewers have seen before is immediately recognized, and exemplars from those new pockets will be pulled as contextual diversity seeds and put in front of reviewers in the very next batch of documents to be reviewed.

 

C. Bottom Line

    1. Majority of documents that had to be reviewed was reduced by more than 90%.

 

 

14.  Case Study: TAR Does Double Duty in a Government Probe Insight Predict Reduces Review and Rescues Privileged Documents

 

A. Challenge: Review Carefully But Control Time and Cost

    Government investigation required the review of 60K documents. Initial sample put richness at 20%.     Review stopped at 60% of total population but got 96% recall. 

 

 

 

B. Using Predict as a Check for Privileged Documents

    1. Documents already ranked for privilege used as training seeds.  Batched top 100 documents that Predict ranked as privileged but reviewers ranked as responsive.  From this batch review team only found 5 privileged documents.   Process continued several more times.  

 

 

   2. Only 500 documents reviewed before no more additional privileged documents, and only seven more privileged documents identified. 

   3. 40% of document production eliminated from linear review 

 

Please reload

Contact Me With Your Litigation Support Questions:

seankevinoshea@hotmail.com

  • Twitter Long Shadow

© 2015 by Sean O'Shea . Proudly created with Wix.com