Personality estimation based on weblog text classification

This paper describes the investigation of personality estimation from Japanese weblog text. Among various personality types, we focus on Egogram, which has been used in Transactional Analysis and is strongly related to the communicative behavior of individuals. Estimation is performed using the Multinomial Naive Bayes classifier with some feature words that are selected based on the information gain. The validity of this approach was evaluated with real weblog text of 551 subjects. The results showed that our approach achieved 12- 25% improvement from baseline. The feature words selected for the estimation are strongly correlated with the characteristics of Egogram.