杏吧原创

Perhaps you could use this to update your online dating profile. A computer model can now predict how we will assess someone鈥檚 personality based on their looks.

We all tend to make snap judgements when meeting people. To find out which facial attributes are most important, of the University of York, UK, and his colleagues analysed 1000 photos of faces, which others had rated on how trustworthy, dominant and sexually attractive they looked 鈥� traits that we are known to assess within 100 milliseconds. The team also measured the attributes of each face, such as the curvature of the lips, the width of the face and whether people were wearing glasses.

Hartley鈥檚 team fed the facial information and the volunteers鈥� ratings into programs that find correlations in data. They found that some attributes, particularly in combination, strongly influence the way faces are judged. Faces with an open, smiling mouth and a wide nose tend to be judged trustworthy, for example. Using this information, the group developed a model that can predict how people will rate a face.

To confirm the correlation, the team reversed the process to see if they could generate computerised faces that looked either trustworthy, dominant or sexy. Then they then asked a different group of people to assess the generated faces.

The team鈥檚 model was able to closely predict how volunteers scored each face in terms of the three traits. 鈥淚f we put all the social impressions on a scale from 0 to 100, we find that, on average, the model鈥檚 predictions differed from the judges鈥� ratings by 9.8 points,鈥� Hartley says.

Participants were consistently better at identifying trustworthy faces than dominant or attractive ones, probably because this trait is largely associated with the size and perceived warmth of a person鈥檚 smile, says Hartley.

Such snap judgements are evolutionarily vital, says Hartley. 鈥淛udging whether someone is a friend or foe is probably an important part of what you have to do when you see a face,鈥� he says. This is worth taking into account when using social media or including a photo with a job application, he says. 鈥淵ou might want to choose the selfie that best reflects the impression you want to create.鈥�

, who works on using avatars to treat hallucinations at University College London, is impressed with the team鈥檚 ability to create faces that elicit predictable responses. He thinks that the model could be used to make avatars more human-like. 鈥淚t represents a step forward in generating realistic static and animated facial images,鈥� he says.

Journal reference: PNAS, DOI: 10.1073/pnas.1409860111