1. Symantec/
  2. Transparent Predictive Coding

Transparent Predictive Coding

Predictive Coding for Dummies
Predictive Coding for Dummies
· Real-world scenarios
· Benefits, challenges and case law decisions
· Step-by-step outline for using predictive coding

Transparent Predictive Coding

Transparent Predictive Coding enables review teams to achieve highly accurate results with minimal input, significantly reducing review time and cost. The solution delivers an unprecedented level of transparency, reducing risk and ensuring the defensibility of the review process.
Key Benefits of Transparent Predictive Coding
  • Reduce review cost by up to 98%
  • Accelerate review timeframes
  • Verify accuracy using built-in statistical sampling
  • Ensure the defensibility of the review process
Download the Transparent Predictive Coding Defensible Workflow White Paper for more details.

Key Features

Transparent Predictive Coding enables review teams to achieve highly accurate results with minimal input, significantly reducing review time and cost. As a form of technology-assisted review, it works by leveraging algorithms that learn tagging criteria from a manually-reviewed training set, enabling users to generate accurate tagging predictions for the remaining documents.

Reviewers benefit from complete visibility into how predictions are generated and the accuracy of predictions across the case. As part of the Clearwell eDiscovery Platform's Review & Production Module, Transparent Predictive Coding is fully integrated with the rest of the eDiscovery process and can be used in conjunction with linear review. The net result is an accurate and defensible review process at significantly reduced time and cost.

Symantec's Transparent Predictive Coding is:
  • Accurate: Produces highly accurate results with minimal human input, further reducing cost and risk.
  • Transparent: Provides visibility into prediction process, so users can make informed decisions with confidence in results.
  • Defensible: Delivers robust, built-in sampling tools and auditing to document results and demonstrate process integrity.

Key Features of Transparent Predictive Coding

Training Source Analytics — Offers sophisticated analytics by criteria such as custodian, discussion, concept, and participant to ensure the selection of highly relevant initial training sets.
Directed Training — Leverages patent-pending active learning technology to automatically suggest each subsequent training set after the initial training iteration for optimal system learning.
Prediction Insight — Automatically provides a prediction score for the document under review, displaying content and metadata relevant to the prediction.
Review Quality Control — Provides a built-in quality control workflow enabling users to assess and improve review accuracy of technology-assisted review.
Advanced Sampling — Provides built-in statistical sampling tools that allow users to select the correct random sample based on the accuracy requirements of the case.
Prediction Templates — Enable users to import and export prediction models to leverage predictive intelligence across cases.

How to Use Technology-Assisted Review

Transparent Predictive Coding uses machine learning technology to learn review criteria using guidance from human reviewers. Once review teams tag documents in a training set, Clearwell eDiscovery Platform identifies tagging criteria common across those documents, enabling it to "predict" the reviewers' tagging decisions for all documents in the case. This technology has been applied successfully in many industries, ranging from filtering email spam to generating personalized recommendations on shopping websites based on an individual’s purchase history. Transparent Predictive Coding is the first application of machine learning that meets unique needs of eDiscovery.
The Transparent Predictive Coding process occurs in three phases:
  1. System Training
  2. Applying Predictions
  3. Measuring Prediction Accuracy
The Transparent Predictive Coding process begins with a set of case documents. Reviewers tag documents in the initial training set based on the unique tagging structure for the case. Clearwell eDiscovery Platform learns from this training set and generates predictions for the rest of the case documents. Reviewers benefit from built-in statistical sampling tools to test the accuracy of the predictions against human decisions.

When to Use Transparent Predictive Coding

Transparent Predictive Coding delivers the flexibility to meet the needs of a variety of different cases based on factors such as budget, timeline, and risk profile of the organization. For instance, organizations can use Transparent Predictive Coding to perform more effective culling, augment the linear review process through batching and ranking, or fully replace linear review with a complete automated workflow. Many types of cases that require review of large numbers of documents can be improved using predictive coding and technology-assisted review, including:
  • Production for legal discovery
  • Government inquiries or investigations
  • Second requests
  • Analyzing productions from opposing counsel
While customers use predictive coding in different ways, here are three common approaches to use Transparent Predictive Coding to streamline the review process:
  1. Identify and produce — Transparent Predictive Coding delivers the ability to complete an entire review workflow with the highest level of cost savings over traditional linear review.
  2. Accelerated Linear Review — At multiple points during the predictive coding process, reviewers can switch to using prediction intelligence for more intelligent linear review. Reviewers can rank documents based on their likely responsiveness to focus on high priority documents, or batch documents to different reviewers based on prediction score.
  3. Intelligent Culling — Before review begins, Prediction Templates can be used to apply prediction models across cases for more effective culling. By leveraging prediction models created for other matters, case administrators can accurately identify and immediately set aside non-responsive items such as junk emails, system messages, and other profoundly irrelevant files.

Is Transparent Predictive Coding Defensible?

As legal teams look to new technologies like predictive coding to enhance the review process, a common question in the legal community is whether these technologies can be used in a defensible manner. If their use is challenged by opposing counsel in court, parties will likely have to explain and defend their approach. To minimize the risk of disagreements, the Sedona Conference Cooperation Proclamation¹ encourages parties to cooperate with opposing counsel during the initial stages of the case and reach an agreement on the methods and technology that will be used. While cooperation reduces the risk of disagreements, parties will likely need to be prepared to explain and defend their approach, in the event they can’t obtain consensus.
In commenting on the defensibility of predictive coding, Judge Andrew Peck² has outlined three key questions that parties should be able to address:
  1. What was done?
  2. Did the process produce defensible results?
  3. Did the process produce "responsive documents with reasonably high recall and high precision?"
Transparent Predictive Coding is the first technology to deliver insight into how predictions are generated, allowing reviewers to make more consistent and accurate review decisions. Using Prediction Insight, reviewers have visibility into why a tag was predicted for the document under review, allowing them to drill down and view document content that supports the prediction. This visibility helps ensure reviewers make accurate decisions and legal teams can defend review workflows that leverage predictive coding. Clearwell eDiscovery Platform also provides a built-in quality control workflow enabling users to measure review accuracy, identify inconsistent tagging, view disagreements between reviewers, and automatically compare predictions and human decisions to assess and improve review accuracy.
Finally, every decision made by reviewers is automatically tracked in an exportable report to demonstrate process integrity to the court. Using Transparent Predictive Coding and technology-assisted review, Symantec customers are taking control of their review process and achieving highly accurate and defensible results.

Symantec Predictive Coding Thought Leadership

Throughout the design and development of Symantec's innovative Transparent Predictive Coding workflow, Symantec developed unique insights regarding technology-assisted review. Below you will find recent videos, articles, and blog posts from Symantec's eDiscovery experts.

Predictive Coding Quick Hit Videos

Predictive Coding Quick Hits Video
Visit the Seventh Circuit Pilot Program’s Resources Page for more videos from the Mock Trial.

Recent Articles

Visit the eDiscovery 2.0 blog for breaking news on predictive coding matters.
  • Twitter
  • Facebook
  • LinkedIn
  • Google+
  • YouTube