Analytics Categorization Sets

Relativity admins can create Analytics Categorization Sets which use example documents to group together conceptually similar documents. Documents may be assigned to more than one category - up to a maximum of five. The admin can limit the number of categories to which a document may be assigned to less than five. When multiple examples are used for a single category, each functions as a query and the categorized documents are the combined hits. No more than 50,000 examples can be used for a single category. At least 5-20 example documents should be assigned to each category. A workspace containing a million or more documents may typically need a few thousand examples for each category.

Documents used as examples should focus on a single concept and have at least two full paragraphs which concern the concept. There should not be irrelevant headers or footers, or a lot of bad OCR text.

The Analytics Categorization Set tab appears under Indexing & Analytics.

A saved search should be created for the documents to be categorized. An Analytics index must be set up before a categorization set is created. The minimum coherence score indicates how similar a document must be in order to be placed in same category as the example document.

A single or multiple choice field can be created to serve as the 'category source', and then individual choices should be added to it for each category. The 'Example Indicator Field' is a Yes/No field. Any documents marked 'Yes' will serve as examples. When 'Auto Synchronize on Categorize All' is set to 'Yes' the value selected for the 'Categories and Example Source' will be used to automatically create categories. Categories and example documents can also be manually added. Text can be entered to serve as an example rather than using a complete document.

Note that when reviewing a document in the Viewer, you have the option to right click on the text and select Analytics Categorization . . . Add Document As Example OR Add Excerpt As Example.

Categories and Examples must be added for a set before the 'Categorize All Documents' option can be activated on the console.

After the documents are categorized, a field will be created with the name of the categorization set, and another field, 'Category Rank' with the numerical rank of each document.

Textual Near Duplicates

A Relativity admin can run a Textual Near Duplicates Summary to get metrics on the number of near duplicates in a document set.

Under Indexing & Analytics, set up a new Structured Analytics Set. Select a document set to analyze, and check off 'Textual near duplicate identification' as the operation.

You can designate the degree of similarity between documents to be regarded as near dupes and also choose whether or not to include numbers in the review.

From the console on the right, run structured analytics.

You'll have the option to analyze only new documents added to the set.

The analysis will be run, and the results will be imported into Relativity

Afterwards, on the right console click, ' View Textual Near Duplicates Summary'

The summary will indicate how many groups of near dupes there are, and also indicate the average degree of similarity.

The 'Total number of documents' indicates the total number of documents in the textual near duplicate groups.

The Kansas Court of Appeals on Tracking Document Review Hours in Relativity

Today, Judge James Vano issued a decision in Ross-Williams v. Bennett, 2018 Kan. App. LEXIS 22 (Kan. Ct. App. Apr. 27, 2018) affirming a lower court decision approving a settlement and reducing the amount of attorney fees and expenses requested as part of the proposed settlement. A law firm used records from Relativity tracking user activity to support the fact that an attorney had worked more than 6800 hours performing document review. The Kansas district court found that the attorney's billing records were dubious, and noted that that, "it takes merely a keystroke of activity once every hour to keep [the computer program] from timing out or logging off a session." Id. at 33.

Plaintiff's counsel submitted a declaration in order to support the time worked by the attorney in Relativity, noting that a session will log out automatically after 62 minutes and used as an exhibit, a log using fields from a Relativity workspace named, 'Total Usage Time', 'Views', 'Distinct Views', 'Edits', 'Distinct Edits', and '# Documents'. This declaration was submitted after the district court's order and so was not considered by the Court of Appeals. The Kansas Court of Appeals ruled that, "simply hitting a key on one's computer every hour does not equate to actually reviewing documents. Accordingly, we do not find that the district court abused its discretion in denying the motion." Id. at 75.

The Relativity Instance Setting Guide, for version 9.5, seems to contradict the assertion that a user will be logged out of a workspace after 62 minutes. See the reference to CookieDuration, which indicates that an admin can set "the length of time that the users are logged in to Relativity before the system automatically logs them off and requires them to re-authenticate" from between 4 and 24 hours. If the court reviews the exhibit submitted with the declaration, they should see metrics relative to how much document review was actually performed. As the Relativity site makes clear, 'Distinct Edits' are, "[t]he total number of documents edited, excluding repeated edits of the same document. The 'distinct' classification is meant to account for duplicate actions performed against unique documents." 'Distinct Views' would show how many documents a user actually looked at.

It appears as though the district court issued its decision, and the Court of Appeals affirmed, on the basis of a report such as this:

. . . when it could have reviewed a report like this: