Computer Program Can Tell What You're Feeling

Thursday, 21 August 2014 - 11:50AM
Technology
Thursday, 21 August 2014 - 11:50AM
Computer Program Can Tell What You're Feeling

Can your computer tell what emotions you're feeling? Bangladesh researchers identified a method that can predict a person's self-reported emotions with up to 87% accuracy.

 

In this study, a computer program used keystroke attributes and text analysis in order to gauge the primary emotion felt by the user. (To be fair, these emotions were painted in very broad strokes; the only choices were joy, fear, anger, sadness, disgust, shame and guilt.) The keystroke attributes included typing speed, "dwell time" (time elapsed between a particular key press and release), and "flight time" (time elapsed between a key release to the next key press). In order to perform text analysis, the researchers provided the program with over 7,500 sample sentences that were associated with a certain emotion. The sample sentences were compiled using data from a survey of nearly 3,000 people from 37 countries, in which they were asked to recount situations in which they had experienced all seven emotions. The program would then choose a primary emotion based on the instances in which the keystroke attributes and text analysis yielded consistent results.

 

Examples of sample sentences:

[Credit: A.F.M. Nazmul Haque Nahin et al]

 

After a certain amount of computer use, the researchers asked the users to report their own emotional states in order to test the accuracy of the program's predictions. The program was able to predict "joy" with the highest success rate at 87%. "Anger" was second at 81%, and "fear" had the lowest success rate of 67%. While previous studies have used both keystroke attributes and text analysis in isolation in order to gauge user emotion, the researchers demonstrated that their combination of the two methods yielded more accurate results than either method used on its own.

 

[Credit: A.F.M. Nazmul Haque Nahin et al]

 

The researchers assert that emotional recognition software could make computer programs more effective in many different arenas, such as gaming, online teaching, and text processing. And, reminiscent of Oxford's eDNA, they believe that it may be able to stymie hackers and identity thieves by detecting feelings of guilt (although we can't imagine that detecting emotions could ever be a reliable method of stopping identity thieves). "A gaming application that can detect and respond to end users' emotions can assess the users' current emotional state and adapt accordingly (i.e. it can change the graphics quality, volume level, control sensitivity level, music selection and many more). Similarly, an emotionally intelligent online system can change its teaching style or contents, change the interface by giving it a more attractive and easy-to-understand look according to a particular student's emotional behaviour."

Science
Artificial Intelligence
Technology

Load Comments