What is Cognitive Analytics and how is it applied in eDiscovery?

In generic terms, cognitive analytics is the process of applying human-like intelligence to unstructured data in order to enrich the value we can extract from it. It’s inspired by the way the human brain processes information, using a combination of experience-driven instinct and learning to join the dots and draw meaningful conclusions.

In eDiscovery terms, cognitive analytics involves applying trained (and trainable) algorithms to unstructured data in order to enrich that data with greater insight and clarity. This ultimately improves human ability for further interpretation.  

There are a plethora of tools and techniques used to achieve this, including data mining, pattern recognition and – one of the more exciting developments in recent times – Natural Language Processing (NLP).

NLP allows inputs to include unstructured data like raw text, handwritten content, email, blog posts, text messages, mobile data and voice transcriptions. It then analyses these communications for much more than just keywords and phrases, looking at context, sentiment and underlying meaning as well. This has massive potential across a huge variety of industries, not least of which is eDiscovery.

Of course, extraordinary potential doesn’t always mean easy accessibility. For a long time, the huge amounts of processing power and storage required for cognitive computing made it all but inaccessible to the vast majority of organisations. With the advent of the Cloud, however, this barrier has fallen away. Cognitive analytics is now readily available at highly competitive price points, fuelling a dramatic increase in the technology’s use cases.

What does that use case look like for eDiscovery? Let’s take a look.

Traditionally, one of the first things parties in a legal matter do is establish a set of “fact patterns”. These “fact patterns” inform the keywords they’ll use to guide their search through the collected dataset to surface results that support or defend their position.

Creating a comprehensive – and accurate – set of fact patterns isn’t easy, however. Particularly if you only have a partial understanding of the facts, or the established facts turn out to be wrong.

That’s where cognitive analytics can be a gamechanger.

Cognitive analytics enables investigators to cast a much wider net than normal, using techniques like sentiment analysis to uncover unusual behaviour patterns or hidden links between people. This could expose leading indicators that would otherwise have been overlooked, providing invaluable insights from which a more comprehensive set of “fact patterns” can be built.

From there, search results can be culled aggressively using pre-trained Artificial Intelligence (AI) models to identify irrelevant content such as SPAM, regardless of volume, without increasing the investigative workload or risk of missing critical details.  Ultimately, this enables investigators and/or legal professionals to surface and disclose keywords or concepts that are more likely to strengthen their position.

As powerful as cognitive analytics is, however, it’s not a silver bullet. It still requires human input to perform at its best.

For example, data must first be in machine-readable format for use by the various supervised and unsupervised AI algorithms, which requires IT skills. Secondly, understanding when a model is mature or stable enough to yield results with confidence is an art in itself. This is best handled by a combination of machine learning experts, data analysts and subject domain experts, who together can collaborate to get the most from the AI technology.

Supervised machine learning algorithms also require human intervention to decipher data they do not yet understand; those edge cases where even two lawyers might have differing opinions. This teaching/learning process is clearly more dependent on  legal experience than IT know-how, reinforcing the ongoing, essential role legal professionals (and investigators) play in the eDiscovery space.