How algorithms discern our mood from what we write online

Dana Mackenzie:

In addition to taking Twitter user’s emotional temperature, researchers are employing sentiment analysis to gauge people’s perceptions of climate change and to test conventional wisdom such as, in music, whether a minor chord is sadder than a major chord (and by how much). Businesses who covet information about customers’ feelings are harnessing sentiment analysis to assess reviews on platforms like Yelp. Some are using it to measure employees’ moods on the internal social networks at work. The technique might also have medical applications, such as identifying depressed people in need of help.

Sentiment analysis is allowing researchers to examine a deluge of data that was previously time-consuming and difficult to collect, let alone study, says Danforth. “In social science we tend to measure things that are easy, like gross domestic product. Happiness is an important thing that is hard to measure.”

Deconstructing the ‘word stew’

You might think the first step in sentiment analysis would be teaching the computer to understand what humans are saying. But that’s one thing that computer scientists cannot do; understanding language is one of the most notoriously difficult problems in artificial intelligence. Yet there are abundant clues to the emotions behind a written text, which computers can recognize even without understanding the meaning of the words.