I think that the Brainspace Discovery 4 analytical engine (http://brainspace.com) is pointing the way to the future of technology-assisted review. In short, Discovery 4 exposes its inferences about which concepts indicate responsiveness and allows reviewers to adjust them.
(NOTE: I have no financial interest in Brainspace. I just think the Discovery 4 product is amazing.)
Here, the reviewer has searched a subset of the Enron collection for the seed concept “california energy shortage.” The Discovery 4 engine shows a set of related concepts, and that there are 44,916 documents related to the concept “california energy shortage”:
Scrolling down the left-side panel, the reviewer sees concepts that are more loosely related to “california energy shortage,” as shown by the fact that the sliders for those concepts are shorter:
By adjusting a slider up or down, the reviewer adjusts the degree of relatedness between the seed concept and the related concept so that it reflects their relatedness for purposes of that search instead of in the abstract. The engine then re-clusters the concepts on the fly and re-ranks the documents based on the revised relatedness weights.
So, for example, if the reviewer drags the slider to increase the weight of a loosely-related concept, such as “electricity situation,” the set of related documents decreases from 44,916 to 95:
If you view one of those 95 results, you can see either an HTML rendering of the document or, as in the next screenshot, a plain text version with key terms highlighted, and a list of the “top terms” in the document:
Of course, searches can be saved and revised.
Brainspace Discovery 4 integrates with predictive coding platforms, both through its own API and through APIs exposed by other products. It integrates with KCura’s Relativity, LEXISNexis’ LAW PreDiscovery application and patent database, and iConect’s Xera, among others.
You can see Brainspace’s concept-tuning technology in action in Avansic’s YouTube video of its Xera platform, incorporating Discovery 4 technology, at http://youtu.be/29gv63gCkHw?t=1m9s. (Note: at the time, Brainspace was named Pure Discovery.)
By allowing reviewers to actually see and directly adjust the engine’s working hypotheses about what features indicate relevance, Brainspace Discovery 4 extends and enriches the kinds of conversations that document reviewers can have with their digital tools.