P2P

12 P E E R T O P E E R : I L T A ' S Q U A R T E R L Y M A G A Z I N E | F A L L 2 0 2 0 email. To identify content that is unique, email threading leverages the header and body information of emails to determine if they belong to the same conversation. The "inclusive" threads are then identified, which are those emails that contain all the unique content in an email conversation, so that noninclusive emails may be ignored. For example, if I send an email to John and Jane and Jane replies to all, several messages are created within the email conversation. But reading only one copy of the email from Jane's reply captures all of the unique content. Depending on the patterns of communication and the proportion of emails in a dataset, email threading may drastically reduce the volume of data that needs to be considered. After employing structured analytics to identify the unique content in a dataset, conceptual analytics can help law departments explore what information is embedded in that unique content. One of the most frequently used conceptual analytics techniques in discovery is called clustering. Clustering is a form of unsupervised machine learning that divides a dataset into groups of conceptually similar documents or clusters. Clustering is especially helpful early in the exploratory process because it requires little or no user input. Imagine a scenario in which I have 100,000 documents from my law department. If I am not already familiar with the dataset, it may be difficult to know where to start in understanding my data. With clustering, documents are now automatically sorted into different conceptual groups. Instead of staring at a list of 100,000 documents, I may have, say, 40 different clusters with labels showing the types of documents they contain. Users can also leverage cluster visualization—which is often presented in the form of circle packs or a clustering wheel—to quickly navigate between groups of conceptually related documents. Not only can I easily choose different clusters to evaluate different concepts in my data, but if I see a cluster that sparks an interest, I can easily drill down further into the subclusters, which focus on the different concepts within the parent cluster. Unlike clustering, sentiment analysis is a conceptual analytics technique with a narrower focus: it aims to capture the emotional signature of a document. With sentiment analysis, users can determine whether a document is expressing positive (e.g., happiness) or negative (e.g., anger) emotions. Such analysis may be used strategically if the documents of interest are correlated with their emotions. For instance, as law departments look to modernize their operations, they may want to identify opportunities for automation. To do so, they may focus on documents containing negative sentiments, since those sentiments tend to be highly correlated with work that is considered repetitive, labor-intensive or prone to human error. Depending on the type of software chosen, sentiment analysis can also capture other types of emotions such as intent, deceit or concern that may bring additional insights about the dataset. In addition to structured and conceptual analytics, predictive analytics and the predictive coding technique F E A T U R E S "Conceptual analytics can help law departments explore what information is embedded in that unique content."

• Cover