Predicting discriminative personality profile of haters from digital texts

Abstract

The rate of malicious behavior in online social media is increasing daily. The increase in malignant social behaviors, especially those involving hate speech, makes it necessary to identify haters from digital texts as quickly and accurately as possible. Recently, many studies have been conducted to identify such behaviors; however, profiling the haters’ personality has not been paid sufficient attention. Retrieving the personality profiles of suspected haters from digital texts is one of the most effective ways to distinguish them from others. This study proposes a novel hatebase-aided personality recognition model that gives more successful results than plain recognition models and predicts the discriminative personality traits of online haters. The proposed model contains the combination of two effective sub-models; a deep neural network model, and a fine-tuned BERT model. While the deep neural network model trained with hate indicators provides an interpretable relation between personality and hate indicators, the fine-tuned BERT model provides relationships between text semantics and personality. Combining these sub-models, the proposed model gives hatebase-related personality recognition results. This study evaluates two popular personality models: the Big Five and the MBTI. According to experiments, compared with other users, online haters are less agreeable regarding the Big Five and fewer thinkers regarding the MBTI.