Error Message

No error message is present.

 

IR uses user-generated Review actions along with certain message metadata (Message Direction, Author, To/Cc/Bcc Recipients, Subject) review and confidence levels (where applicable) to Intelligently Review and Mark Random Sampled items and items found via Searches where IR is enabled. The boundary for this IR Learning information is the Department, meaning Departments cannot share the Learning information and one Department will not use the IR Learning information from another Department.

Note: IR will only work if Reviewers Mark items (found by Random Sampling or Searches) as Reviewed Relevant or Reviewed Irrelevant. IR will not automatically make its own presumptions on what should be Relevant or Irrelevant - it relies on the Reviewers having Marked items as Reviewed Relevant or Reviewed Irrelevant first so it can learn from those items. IR may or may not be enabled at the Department level, however, this does not affect the Learning behaviour. As long as items are Marked as Reviewed Relevant or Reviewed Irrelevant, the Learning information for those items is used. It is also important to note the Learning information is only updated when Random Sampling runs. Thus, if Random Sampling has already run today, then items Marked today will not have their Learning information be updated for the Department until the next Random Sampling run occurs tomorrow.

For Searches, there are currently 3 Learning settings:

- None: No Intelligent Review enabled.
- Search and prioritize: Provides a mix of Relevant and Irrelevant items without favouring a higher percentage of Relevant over Irrelevant or Irrelevant over Relevant.
- Search and then Sample only relevant content: Adds only Relevant items to meet the Sampling percentages. If not enough Relevant items are available to meet the Sampling percentages, no more Irrelevant items are added.

If a Search was configured to use the 'Search and then Sample only relevant content' IR setting, this means the Search will first find all items that meet the Criteria, then apply any Learning information to the hits to include items that match the Learning behaviour for Relevant items, finally would apply the Sampling percentage as configured in the Search.

Therefore it is possible a Search's Hits and Sampled columns may not have the expected numbers when using the 'Search and then Sample only relevant content' IR setting. If the IR process found only half the items found by the Search matched the Learning behaviour criteria for Relevancy, it would process that number of items and them apply any Sampling percentages.

Using the examples above:

- If the Search is set to 100% Sampling, has the IR setting 'Search and then Sample only relevant content' enabled, finds 1000 hits and only half of them match the Learning behaviour criteria for Relevancy, then it is expected the Sampled number of hits should be 500.
- If the Search is set to 10% Sampling, has the IR setting 'Search and then Sample only relevant content' enabled, finds 1000 hits and only half of them match the Learning behaviour criteria for Relevancy, then it is expected the Sampled number of hits should be 50.

If the IR process found that none of the items met the Learning behaviour criteria for Relevancy, it would not show any Sampled items. Thus the Hits column for such a Search would show the number of hits found, but the Sampled column would show no hits (0).
 

 

Per the information above, this behaviour is expected. If a Search is configured to use IR, then the Search processing will take into account the Learning information for the Department and apply it to the hits, based on the IR setting selected.

A search can be run without IR enabled, i.e., IR set to None, so all hits found by the Search may be added to the Review Set upon Search Acceptance.

When running Compliance Accelerator (CA) Searches, the number of hits and number of Sampled hits may be seen in the Searches tab under the Hits and Sampled columns respectively. For example, if the Search is set to 100% Sampling and it finds 1000 hits, it is expected the Sampled number of hits should be 1000. Another example would be if the Search is set to 10% Sampling and it finds 1000 hits, it is expected the Sampled number of hits should be 100. When Searches are run with Intelligent Review (IR) enabled, there is a possibility the number Sampled is less than the calculated hits multiplied by the Sampling percentage. In some cases, the Sampled column may even show no hits (0).