What we’re thinking about

Insights, news, and tips from our top tech and business innovators.

Review half a million documents between coffee and lunch

Annelore van der Lint | September 12, 2018

One of the key highlights of our recently introduced ZyLAB ONE 5.0 is the Assisted Review option that helps reviewers to quickly find relevant/responsive documents.

Assisted Review is a process in which after a brief training, the computer helps to identify potential relevant/responsive documents. Supported by advanced machine learning, a component of artificial intelligence (AI), the computer uses a few relevant and responsive samples and calculates which other documents are of similar content. This speeds up the review process and improves the quality of the review.

 

Finding Documents: Hands-on Workshop with Machine Learning

Recently we organized a Friday-morning workshop for lawyers to let them experience ZyLAB ONE and to use Assisted Review to find documents. At this workshop, we divided the group in several review teams and asked them to find all relevant and responsive documents related to travel, using standard search techniques. Then we focused on finding actual trips or travel arrangements and to exclude advertisements or travel offers. For this exercise, we used the Enron data set with about 600,000 documents.

After trying to find these types of documents, the review teams found out that it was pretty difficult to create a search query that identifies these specific type of travel documents. Most searches that were used, returned too many documents (too much recall).

We then started the actual Assisted Review project. We reviewed a set of 40 documents created based on a search with a few relevant/responsive documents. After that, the Assisted Review system used the relevant/responsive documents to train itself and it offered to the reviewers a new review batch. To the reviewers’ surprise, this review set (we call them training rounds) contained already more relevant/responsive documents.

This review set was also processed by Assisted Review and so another review set was created. Again this set contained relevant/responsive documents; even more than the previous training round. After a few rounds, the lawyers were convinced they had found all relevant/responsive documents.

 

Resources 

It is great to see this hands-on workshop really works well to demonstrate the capabilities of technology such as machine learning. Instead of trying to illustrate the benefits via dull PowerPoint slides, participants now firsthand experience the power of the best algorithms and most advanced data science techniques of ZyLAB’s Assisted Review.  As one of the participants said; “Reviewed half a million documents between coffee and lunch during a demo at ZyLAB.”

Discover how easy it is to get an assisted review project up and running in this quick start tutorial