Cluster Visualization QC Workflow

Improve the efficiency and effectiveness of your quality control workflow by using cluster visualization to help identify potential coding discrepancies. This recipe describes how you can use cluster visualization to identify patterns and visually compare groups of conceptually similar documents to uncover potential coding inconsistencies.

Requirements

  • Relativity 9.2 or above
  • An active Analytics index
  • An existing cluster

Directions

This recipe involves a scenario where the case team has performed a first-level review for responsiveness. Cluster visualization can help QC the responsiveness coding. The approach described here could also be used to QC privilege coding or issue coding.

  1. Ensure that the reviewed and coded documents are included in an existing cluster set.
  2. From the Documents tab, select the cluster set in the cluster browser and click Visualize Cluster.
  3. In the search panel, click Add Condition and select the responsiveness coding field. Create a condition that will return documents coded as responsive. Click Apply, then click Run Search.
  4. The system displays a heat map overlay on the cluster visualization that indicates the percentage of documents coded as responsive in each cluster. The darker the shading, the higher the concentration of responsive documents in the cluster.
  5. Throughout this scenario, you can focus in on the documents in a cluster by right clicking the cluster and clicking Select, then Apply. You can select multiple clusters by doing this for each cluster you want to focus on.
  6. The main cluster visualization view (the circle pack view) provides the following visual clues to aid in your QC review:
    1. Darker clusters – The darker shading indicates that a cluster contains a high percentage of documents coded as responsive, meaning that the documents in the cluster have been coded consistently. Since documents within a cluster are conceptually similar, we would generally expect them to be treated similarly, and this is what the darker shading tells us.
    2. Lighter clusters – The lighter shading indicates that only a small percentage of documents in the cluster have been coded as responsive. The documents in these lighter shaded clusters would warrant further investigation to determine why only a small portion of conceptually similar documents were coded as responsive.
  7. Continue your QC review by right clicking a darker shaded cluster (one with a high percentage of responsive documents) and selecting View Nearby Clusters from the right click menu.

  8. The Nearby Clusters visualization places the selected cluster in the middle of the screen and shows you other clusters that are conceptually similar to it. The closer a cluster is to the center cluster, the more conceptually similar it is.

    Along with the circle pack, the clusters can be visualized on the dial by right clicking and selecting View Dial.

  9. The Nearby Clusters visualization reveals the following additional insights to help guide your QC review:
    1. Darker clusters – We would expect clusters that are nearby, or conceptually similar to, the center cluster to also have a darker shading, indicating that they too contain a high percentage of documents coded as responsive.
    2. Lighter clusters - Clusters with a lighter shading would probably warrant additional investigation. The system is telling us that these clusters are conceptually similar to the dark cluster in the middle (which contains a high percentage of responsive documents). However, only a small portion of the documents in these lighter clusters have been coded as responsive; we might want to investigate why this is. A cluster that is completely white contains no responsive documents, yet is conceptually similar our center cluster, and should also be investigated.
  10. Note that with the Responsiveness filter set and active, only documents coded as responsive are returned. If you would like to access all the documents in a selected cluster, simply clear the box on the filter card, click Run Search, and all documents within a cluster will be returned.

References