Abstract:
An increasing number of users of social networks has significantly contributed to a enor increase in the amount of information on the Internet. Often, the content these users post on social networks can provide valuable information about their personality (for example, in terms of predicting job satisfaction, certain preferences, and the success of professional and romantic relationships). Personality prediction, involves extracting digital content into traits and mapping them onto a personality model. Because of its simplicity and proven capabilities, a well-known model of personality called the Big Five personality traits has often been adopted in the literature as the standard for personality assessment. To date, many algorithms can be used to extract embedded contextualized words from text data for personality prediction systems; some of them are based on ensemble models and deep learning. While useful, existing algorithms like LSTM and RNN suffer from the following limitations. First, these algorithms take a long time to train the model due to their sequential inputs. Second, these algorithms also lack the ability to capture the true (semantic) meaning of words; therefore, the context is easily lost. To address these limitations, this paper presents a new prediction method that combines a deep learning architecture with pre-trained language models such as BERT, GloVe, and TF-IDF, BOW, and NLP sentiment. Finally, the system makes the decision based on the average of the model to make predictions. Unlike previous work using only two social network data sources with open and closed vocabulary extraction methods, the proposed work uses multiple sources of social network data, e.g. extraction methods. Our experience with the developed method has been encouraging, as it outperformed similar works in the literature. Specifically, our results achieve a maximum accuracy of 89.2% and a metric score of 91% F1 on the Facebook dataset.
