Google Researching Real-Time Frustration Detection

Top 10 Most Hated Tech Mascots Ever

Over the past few years, the standard opening gambit for launching a company that competes with Google has been to state that keyword search doesn't work very well and to promise something better.

While Google won't dismiss its service quite so readily, its executives and engineers acknowledge that online search could be better and regularly deliver improvements, such as the recent introduction of Google Instant.

In fact, search does work pretty well. According to metrics firm Hitwise, the success rate for searches is around 75% and is pretty much the same for Google and Bing. But there's room for improvement.

Usually, these improvements involve implementing sophisticated ways to figure out what the user is thinking, because bad search output often follows from inadequate or incoherent user input. The hope is that Google's machine smarts can determine user intent when that intent is not spelled out.

But figuring out what the user is feeling also turns out to have implications for user happiness and search success. So Google's researchers have been studying how users behave when they're having a hard time conducting a successful search for information.

Like bad poker players, frustrated searchers have tells, behaviors that betray their thoughts. They may frown or lean closer to their monitor, to make sure they're not missing anything.

"[A]fter a couple of unsuccessful searches, we started noticing interesting changes in behavior," wrote Google user experience researchers Anne Aula, Rehan Khan, and Zhiwei Guan in a blog post last week. "In addition to many of them sighing or starting to bite their nails, users sometimes started to type their searches as natural language questions, they sometimes spent a very long time simply staring at the results page, and they sometimes completely changed their approach to the task."

And because these signals can be detected, it's conceivable they could be detected by computers, were users to somehow warm to the idea of Google watching them through a Web cam.

"[W]e believe we can use [behavioral observations and other metrics] to build a model that will one day make it possible for computers to detect frustration in real time," the researchers said. Think of it as another form of the voice-stress monitoring employed by some call centers.

In theory, frustration detection could be done today. A mobile device with an accelerometer should be able to determine when a furious user has hurled it out the window, assuming it happens to be running an appropriate background process. Perhaps it could even respond appropriately: "I'm sorry, Dave, but you can't throw me like that."

However, Google is aiming to capture more subtle changes of emotion prior to that point, so it can make your search experience a happy one.