Using Deep Neural Networks Architectures to Identify Narcissistic Personality Traits

Personality is the characteristics of a person represented by thoughts, feelings and behaviours in a certain way. Knowing the personality characteristics of an individual can help improve interpersonal relationships, regardless of their type. Virtual media of social interaction is a rich source of information where online users share and post comments, and express their feelings of likes or dislikes. This information reveals traits about the personality and behaviour of users. In this sense, it is possible to identify personality traits of the dark triad through computational models. In this area, research has found correlations between personality traits and users' online behaviour. In this study, we propose a computational model that uses Neural Network Architectures and Transformer models to identify narcissistic personality traits in Spanish-language text based on the Narcissistic Personality Inventory (NPI) test. Specifically, we leverage the ability of the pre-trained Transformers models BERT, RoBERTa and DistilBERT, to capture the semantic context and structural features of text using sentence-level embeddings. These attributes make them suitable for multi-class classification tasks, such as identifying personality traits from reviews. Furthermore, the model utilises the algorithms Glove, FastText, and Word2Vec to generate embedding, which are used to represent vectors of semantic and syntactic features of words in narcissistic expressions. The semantic information is then used by several neural network architectures—namely SimpleRNN, LSTM, GRU, BiLSTM, CNN + BiLSTM, and CNN + GRU—to construct a multi-class model for automatically identifying narcissistic personality traits. The model's performance is assessed using a Twitter dataset that has been annotated by psychology experts and increased using augmentation techniques such as Back Translation, Paraphrasing, and substituting words with their synonyms. Ultimately, the results indicate that BERT and RoBERTa Transformers yield better accuracy and precision compared to Neural Network Architectures.