Our emails are a dead giveaway. The words we use in the messages we send can reveal not just our gender but also our emotions and maybe even our personality traits.
Saif Mohammad and colleagues from the National Research Council Canada, used sentiment analysis to uncover the feelings buried inside email. "It's an efficient way of generating data about the emotional content of huge amounts of text," says Mohammad. "There's been a lot of research based on positive and negative emotion, but with all this data available it makes sense to understand what we can learn from all the emotions."
The team created a huge "sentiment" database using crowdsourcing by hiring workers for Amazon's Mechanical Turk to pair 24,200 words with emotions – "ice cream" paired with "joy", for instance, and "gardening" with "peace". Comparing the words used in a corpus of 32,045 emails made public after the Enron scandal to this database, made it possible to assess emotional tone and how it varied with gender.
The results showed a marked difference between sexes. Women had a tendency to use words relating to joy or sadness, while men favoured those associated with fear or trust. Both sexes used more joyous and cheerful words when writing to women, and men used more anticipatory language when writing to women, such as "prepare" or "hope".
The team also tried to discern personality traits from emotional content. "If you're angry occasionally, that's fine," explains Mohammad. "But if you're angry all the time, it's part of your personality." Armed with a different database of 585 emotions and associated words generated from tweets and hashtags, the team taught an algorithm to try and identify personality types from a short piece of written text.
Thousands of samples – short stream of consciousness essays, each judged by psychologists to represent one of the Big Five personality traits of extroversion, neuroticism, agreeability, conscientiousness or openness – were used to train the algorithm to match combinations of emotions to traits. When tested, its analyses of unseen essays agreed with the psychologists 99 per cent of the time.
Predicting personality traits and gender of an otherwise anonymous individual could help spot warning signs of depression, the researchers claim, or even help with forensic analysis of online crime. But there's an even bigger picture, too. "If you wanted to know about how the latest iPhone release was received, deep emotional analysis of social media could provide incredibly rich insight," says Mohammad.
Further work is required before the findings can be used more widely. "Overall it's very impressive and interesting research," says Mike Thelwall, at the University of Wolverhampton, UK. But he cautions that because the email data belonged to a very specific group of people, the results may not apply more widely.
Journal References: arxiv.org/abs/1309.6352v1, arxiv.org/abs/1309.6347v1
If you would like to reuse any content from New Scientist, either in print or online, please contact the syndication department first for permission. New Scientist does not own rights to photos, but there are a variety of licensing options available for use of articles and graphics we own the copyright to.
Have your say
Only subscribers may leave comments on this article. Please log in.
Only personal subscribers may leave comments on this article
All comments should respect the New Scientist House Rules. If you think a particular comment breaks these rules then please use the "Report" link in that comment to report it to us.
If you are having a technical problem posting a comment, please contact technical support.
