Currently there is a plethora of personality models in use in Psychology. Arguably, the dominant model is the so called Big Five model, which identifies five traits that compose human personality. With the rise of computers in the past decades, automatic personality assessment models have been gaining popularity. Additionally, with the rise of social media, there is a large amount of data containing information on the personality of the users. Regarding text-based personality prediction, large pre-trained language models have been gaining popularity, achieving state-of-the-art results in multiple cases. In this work, two datasets containing textual data, from the Facebook and Twitter social media platforms, were used to predict the Big Five personality traits. The proposed model resorts to the large pre-trained language model, Sentence-BERT, to extract sentence embeddings and a neural network as a regression model. The results obtained with the Twitter dataset outperformed state-of-the-art by around 16–18%, though with the Facebook dataset underperformed when compared with the state-of-the-art. Additionally, a comparative analysis was performed of how data from different sources can be combined and applied to one another, in the scope of personality trait prediction.