Predictive Coding Workflow

Create a predictive coding workflow to index, seed, and train documents. The workflow of predictive coding is as follows:

  1. Create a predictive coding project (the container for the document set, the index field choices, and the classifiers used to create the predictive coding index).

    2. Note: Multiple coding profiles can be created for a single project, and coding profiles can be applied across multiple projects.

  2. Create a seed set of documents within the document set that predictive coding can learn from by either:
    1. Selecting a random document set and begin an initial review

        2. -or-

    2. Selecting a partially coded document set where the coding adhered to the four corners rule of contract law. 

    3. Note: Predictive coding requires documents to be coded on an individual basis, not using family or email threading coding decisions.

  3. Continue reviewing a document set selected as a batch or part of workflow.
  4. Run a training session to generate scores for the documents in the coding profile document set (both coded and uncoded).

    5. Control Sets enable you to assess the performance of trained classifiers.

    You can view a list of all training sessions that have been conducted for the project. This training history provides information about the start and end times of the task, task status, index size, and training population at the time of task, as well as any document-level exceptions that may have occurred.

    Note: Predictive coding scores can be viewed on the Review Document List..

  5. Continue until the linear review exhausts responsive documents.
  6. Quality control the remaining pool. KLDiscovery's EDiscovery Technologists can create a random sample set to determine the percent, or number, of remaining responsive documents.
  7. Decide to continue or stop reviewing.