Leveraging NLP for Personality Prediction from Social Media Text

The ever-increasing nature of data generated by social media offers ample reasons for the extraction of meaningful information and proper categorization of the same information. One major challenge in natural language processing is text classification, as it allows the organizing of unstructured texts into categories of interest for value addition, such as sentiment. [1] Although deep learning models have brought so much breakthrough in this field, accuracy in classification is still a viable space for improvement. This study, therefore, applies the natural language processing techniques on a dataset from a WhatsApp group to determine the sentiment using five models of deep learning: Neural Network, Recurrent Neural Network, Long Short-Term Memory, Bidirectional LSTM, and Convolutional Neural Network. We have proposed a hybrid approach of CNN- BiLSTM that uses activations, dropouts, filters, kernel sizes, and numerous layers along with feature extraction to enhance the classification accuracy of sentiment. The proposed model’s results are tested with comparisons to earlier studies. Among the individual models used, the highest result was 81 percent by LSTM and BiLSTM, and this hybrid model was much better as it achieved an accuracy of 88 percent. In fact, the proposed model can be seen to perform well in comparison with existing work concerning the task of sentiment classification..