Can you find out if  someone you know a psychopath by reading their tweet?

Predicting Dark Triad Personality Traits from Twitter usage and a linguistic analysis of Tweets

Chris Sumner

 Alison Byers

Rachel Boochever

Gregory. J. Park

Presented By -

Alaukik

Ameya

Prediction of personality traits can be performed in two ways, both of which were examined in this paper:

• A classification task, where the goal is to identify individuals with particularly high or low values of a trait according to some predetermined cut-off.

• A regression task, where the goal is to predict an individual’s score for each of the eight personality traits based on their Twitter usage.

•  Support Vector Machine (SVM)
•  Decision tree algorithm.
• Naïve Bayes (NB) classifier.

Narcissistic traits were significantly positively correlated with ‘other punctuation’ (OtherP), which includes the @ and # characters (r(2,614) = 0.073, p = < 0.001). The @ and # characters have special significance when used in Twitter. The @ character is used before other characters to signify a Twitter username and is typically used in replies and Tweets which mention other users, while the # character indicates a “hashtag”,

Machiavellian traits were significantly positively correlated with swear words (r(2,614) = 0.129, p = < 0.001), anger (r(2,614) = 0.116, p = < 0.001) and negative emotions (negemo) (r(2,614) = 0.073, p = < 0.001), suggesting that as levels of Machiavellianism increase, so does the use of negative and hostile language. Machiavellian traits were significantly negatively correlated with positive emotion (r(2,614) = -0.118, p = < 0.001) and the use of the word “we” (r(2,614) = -0.070, p = < 0.001), showing that as levels of Machiavellianism increase, references to other people, i.e. “we”, decreases.

Psychopathic traits were significantly positively correlated with swear words (r(2,614) = 0.187, p = < 0.001), anger (r(2,614) = 0.151, p = < .001), death (r(2,614) = 0.094, p = < 0.001) and negative emotion (negemo) (r(2,614) = 0.084, p = < 0.001). We also saw significantly positive correlations between

psychopathic traits and filler words (r(2,614) = 0.073, p = < 0.001).

In no cases did all three Dark Triad traits share statistically significant results. For example, while we saw increased levels of swearing and anger in relation to Machiavellianism and psychopathy, these relationships did not appear with narcissism.

Conclusion

This study highlights that there are relationships between Twitter activity, Dark Triad and Big Five personality traits, but that the practical performance of machine prediction is currently poor when applied directly to an individual.

The results demonstrate that while our models display a high degree of accuracy, defined as (TP + TN) / (TP + TN + FP + FN); the TPR and TNR remain poor, highlighting a need for greater focus on evaluation criteria in future studies.