Conceptual Index Updates

When working with a conceptual index for Relativity analytics, there are different circumstances when the training set, the searchable set, and the index should be updated.

• When documents with new concepts are added, these documents should be added to both the training set [from which term relationships are learned] and the searchable set [on which the analytics tools like clustering and categorization are run]. Perform an incremental build on the index.

• When documents with no new concepts are added, these documents should be added to the searchable set. Perform an incremental build on the index.

• When documents that make up more than 10-30% of the entire set are added, these documents should be added to both the training set and the searchable set . Perform an incremental build on the index.

• When documents that make up less than 10-30% of the entire set are added, these documents should be added to the searchable set. Perform an incremental build on the index.

• When documents are removed from a training or searchable set, an incremental index build should be performed.

• When new words are added to the list of concept stop words, the index should be deactivated and then rebuilt.

• When extracted text is modified, perform a full build on the index.

• When email header and repeated content filters are revised, perform a full build on the index.

Email thread ID node / indented segment mismatch

When email threading is performed in Relativity, the analytics server can detect more segments than there are blocks in an Email threading ID. This will be an indication that the latest email matches an earlier segment. If the ID has more blocks than detected segments, the segments may have been edited or corrupted.

Segments are calculated by the number of indentation levels. Relativity uses indentation squares like this to show the number of indentation levels.

When a server applies a confidentiality footer, it can cause the email threading ID to get an extra block. So even though there would in fact be three segments in an email the ID would look like this: F00000abc-0000- 0000+0000+.

Random Sampling in Relativity

By default Relativity enables sampling on the document object. If an admin has kept this setting in a workspace you should see an icon that looks like this:

. . . to the right of the View bar. When you click on this object, you'll get options to create three different types of randomized samples.

1. Fixed size - will create a random sample set of a specific number of documents.

2. Percentage - will create a random sample set of a particular percentage of documents.

3. Statistical - will create a random sample based on how many documents must be included in order to get an accurate representation of the data set.

The sample will be based on the documents included in a folder, view, or saved search that you have selected.

After selecting one of these radio buttons, a document list will be generated, and the setting will be shown in a title bar:

Note that Relativity does not recommend using statistical sampling on sets of fewer than 300 documents. Confidence levels of 90, 95, and 99% are available for statistical sampling. The margin of error can also be adjusted. If you work with a larger document set, it may make sense to lower the margin of error value. To be clear, the margin of error indicates by how many points the sample may differ from the full set. The confidence level indicates how often the sample will fall within that margin of error. So a statistical sample with a confidence level of 90% and a margin of error of 4% will be within 4% of the values of the full set 90% of time.

Sampling will not be an option when a view is set to show the related items for each document family.