Field for Responsive Designations in an Active Learning Project Can Have More Than Two Choices

Note that in the current version of Relativity, when you are setting up an Active Learning field, you are no longer limited to selecting a Review Field (the field for responsive and non-responsive coding) that only has two possible choices. An update made in 2022 now permits you to allow a user to select from more than two choices as options for this field. The field type must be set to 'single choice', not 'multiple choice' . . .

. . . but you can add more than two choices to be the option for the single selected choice in the field:

On the Relativity Analytics Study Resources page, the training video on Administering an Active Learning Project , which is listed in the current version of the Relativity Analytics Specialist Study Resources guide as a resource still references the prior version and will instruct you that the Review field for Active Learning can only have two choices.

. . . and the Assisted Review Active Learning Guide for Server 2022, dated April 26, 2023 states that:

. . . BUT the Relativity Help page on Creating an Active Learning project states that:

Active Learning requires a single choice field with at least two choices for reviewers to code on. One choice must represent the positive/responsive designation, and another must represent the negative/not responsive designation. Any additional choices will be considered neutral, and they will not be used to train the Active Learning model.

See: https://help.relativity.com/RelativityOne/Content/Relativity/Active_Learning/Creating_an_Active_Learning_project.htm

I raised this discrepancy with Relativity support this week and they confirmed that the training materials are outdated and that you can indeed have a 'neutral' choice linked to the Review Field selected for an Active Learning project.

Expanding on keyword expansion

Don't miss that when you are running keyword expansion from the Document tab in Relativity, you can expand on a kewyord in the in conceptual keywords expansion window.

After selecting a conceptual analytics index when running an index search, an Expand link will appear in the search window:

If you click on Expand, the keyword expansion window will open. When you enter a term in this window, assuming the term is included in the index's training data source, Relativity will show other terms correlated to your term in its spatial index. Words with a coherence score of 50 or greater may be returned.

In this example the keyword 'enron' is shown to be closely related to five other terms within the text of the documents used to train the analytics index. When we click on the word 'baxter' in this list, the keyword expansion list will update to show the keyword terms that are correlated to in the analytics index.

You can check off terms and then use the 'Copy Selected Keywords' button at the bottom of the window to generate a Boolean search to use in an Index Search.

You can add or remove filters on the search results. If you want to filter by a number range in the Rank field, use >= OR <= . > OR < by themselves will result in a syntax error.

This screen grab shows the process in action.

set regex filters in Relativity structured analytics sets for pipeline text

Keep in mind that when you set a regular expression filter for a structured analytics set in Relativity, the regex filter will not be run against the extracted text as you can see it for a document in the Viewer. While the extracted text is displayed with line breaks and whitespace, this text is transformed when it is in the Analytics pipeline. The pipeline text uses the regex \r\n markers [return and newline] in place of line breaks, and will consolidate multiple blank spaces to a single space.

Extracted text may look like this:

. . . but the pipeline text will look like this:

There's just one long string of text in the analytics pipeline. So if you want to search for an email footer reading, "Under the General Data Protection Regulation GDPR 2016 679 we have a legal duty to protect any information we collect from you", accounting for varying GDPR sections, a regex filter for a structured analytics set . . .

. . . should be set like this:

. . . without the additional spaces before and after the relevant GDPR section.

If you want to filter out multiple disclaimers added to email footers, and Bates numbers from more than one party's production, you'll need to craft a single regex search which can account for all these targeted terms. No more than one regex filter can be applied to a structured analytics set.