Predicting the Big Five for social network users using their personality characteristics

Abstract

An enormous amount of data in the form of posts, images, and social network activities resulted from the intensive usage of social networks. Many researches proved the presence of relationships between features extracted from these data and the Big Five personality traits. This motivated scientists to implement different approaches to infer personality traits using users' publicly available data. None of the existing studies examined psychological personality characteristics concealed in the user's generated text as predicting factors for the traits. This paper presents a new approach for personality traits assessment using the personal Values and human Needs personality characteristic features extracted from users' generated text on social media. For predicting the Big Five classes, the proposed features are employed, single and combined, with various supervised machine learning techniques. For adjusting trait binary classification, two marginal thresholds are tested: the median-based split and the third quartile split. Experimental results showed that regression models using the proposed Values and Needs personality characteristic features can classify the Big Five personality traits with an accuracy ranging from 72% to 79%. Furthermore, the proposed approach surpassed the existing related work that employs traditional textual features by 11.7%.