Researchers from Northwestern University and Microsoft Research developed machine learning algorithms that predict a user’s comfort with an email notification depending on a given situational context.
Today, people may receive anywhere from dozens to hundreds of notifications per day across a variety of connected devices. Contingent on the time of day that a notification is received, a user could be surrounded by other people who inevitably become privy to any information disclosed by the notification. UX Research and Strategy professional Susan Farrell argues that this ever-present connection makes us vulnerable to “computer-assisted embarrassment.” The band-aid solution: current devices usually allow for notifications to be turned on or turned off, or for previews to be shown or not shown. A more intelligent system, however, would avoid the pitfalls of this one-size-fits-all approach, displaying important or time-critical information when appropriate while preserving the privacy of less contextually suited emails. For instance, the typical employee might want an email from their doctor to be treated differently than an email regarding a timely business development while they’re on the clock.
Before designing any machine learning algorithms, the researchers administered two studies in an enterprise environment. They sought to gain awareness of how email notifications relate to information disclosure and how users perceive information disclosure risks based on context. The first study consisted of an exploratory survey where users reflected on recent emails and meetings; the second study collected contextual email data in which email-meeting pairs were labeled and then features for each pair were extracted. A total of 1,040 email-meeting pairs from 169 participants were amassed. From there, the researchers used the collected data to find feature sets and algorithms that anticipate information disclosure risk. All three classifiers performed better than chance at predicting the degree of user comfort when revealing a particular email notification in a given meeting.
Since the experiment occurred in a specific U.S. information technology company, the researchers caution that their findings, however promising, may not be entirely generalizable. Information disclosure risk can vary widely between industries, companies, and cultures. Nonetheless, the features implemented in the classifiers can find widespread utility when attempting to reproduce these results in other settings. Future studies undertaken by the researchers may examine other types of notifications, sender disclosure risk, and personal email-context pairs.
Find out more here.
I currently study Computational Linguistics as an M.S. candidate at Brandeis University. I received my Bachelor's degree from the University of Pittsburgh where I explored linguistics, computer science, and nonfiction writing. I'm interested in the crossroads where language and technology meet.