DENVER, Aug. 2, 2018 /PRNewswire/ — Organizations can dramatically reduce the time and effort associated with the costliest part of electronic discovery, document review, by up to 88.5 percent using technology assisted review (TAR) based on a continuous active learning (CAL) protocol, an extensive simulation by Catalyst found. The simulation found that both high efficiency and accuracy are achievable using virtually any document as a seed to begin training and initiate the CAL ranking.
Catalyst conducted the simulation for a Big Four accounting firm to demonstrate the effectiveness of Insight Predict, Catalyst’s TAR technology based on CAL. The client had manually reviewed 5,000 documents and found 55 to be responsive. The document collection would have been challenging for TAR 1.0 systems: it was small, had a low prevalence of responsive material, was in Japanese, and 20% of the responsive documents were hard copy documents that had to be OCR’d (optical character recognition).
Catalyst ran 57 simulations, each with a different starting seed document, keeping every other aspect of the simulated review constant. The key findings include:
- Across all simulations, Insight Predict found 75% of the responsive documents (recall) after reviewing 11.5% of the collection (88.5% savings) and achieved 100% recall after reviewing just 31.1% of the collection (68.9% savings).
- Regardless f the seed document used to start the process (including relevant documents, non-relevant documents and a synthetic seed), results across all simulations were nearly identical.
- Predict performs well on low richness, OCR and non-English document collections.
“These experiments demonstrate the real, significant savings of using continuous active learning,” said Tom Gricks, managing director of professional services and one of the leads for the simulation. “There’s still debate about the extent to which TAR reduces the time and cost of review. We’ve conducted hundreds of TAR reviews for our clients, and the findings in these types of simulations underscore what our clients have experienced—continuous active learning saves an enormous amount of time and money and increases efficiency of review.”
“CAL’s performance is dramatically superior to TAR 1.0,” added Dr. Jeremy Pickens, chief scientist, Catalyst. “Because TAR 1.0 relies on inducing a static model of responsiveness, it often requires one first have a high volume of documents, and within that, a fair amount of relevant material. Unlike TAR 1.0, with CAL there’s no need for a training phase followed by a review phase. The algorithm continuously improves over the course of the review, boosting speed and savings.”
Download the case study here.
Catalyst builds, hosts and supports the world’s fastest and most powerful e-discovery platform. For 20 years, Catalyst has helped global corporations reduce the total cost of discovery and take control of complex, large-scale discovery and regulatory compliance. To learn more, visit catalystsecure.com or follow the company on Twitter at @CatalystSecure.
View original content with multimedia:http://www.prnewswire.com/news-releases/new-study-involving-57-simulations-shows-the-simplicity-effectiveness-and-savings-achievable-with-continuous-active-learning-300690601.html
Source link http://www.tmcnet.com/usubmit/2018/08/02/8796164.htm