Words Mark the Nerds: Computational Models of Personality Recognition through Language

User models in human-computer interfaces have focused on various user characteristics, but there is little user modeling of the most fundamental dimension of variation between humans: personality traits. We explore here the possibility of automatically acquiring such models by simply observing the user’s language. We automatically learn models for personality recognition from different corpora and sources of personality evaluation. The models are completely transparent, i.e. they can run in the background evaluating the user at every conversational turn, and provide input for the system to adapt to the user. Results show that recognition models based on observed personality perform significantly better than a baseline of the average personality score, as well as better than models using self-reports. An analysis of decision trees confirms previous findings linking language and personality, while revealing many new linguistic markers.