Pre-trained Word Embeddings In Deep Multi-label Personality Classification Of YouTube Transliterations

Abstract

The study of personality computation involves util-ising algorithms to analyse differences in behaviour and person-ality. Numerous applications, including human resource management, criminal profiling, and human-computer interaction, could receive help from automatic personality analysis. Psychology enables us to analyse personality in terms of traits, which are adjectival indicators. A MLC problem arises because it is possible for a single person to display several traits and not just one. Therefore, traditional machine learning methods might not be effective. In this study, transliterations from the YouTube Personality dataset were utilised to classify personalities using multi-label semi-supervised learning algorithms. The supervised algorithms that we used are; Neural network, CNN, RNN, LSTM, Bi-LSTM, GRU, and Bi-GRU. In our experiments, we trained all the algorithms with Keras, GloVe, FastText and Word2Vec embeddings. The results are very promising with F-score of up to 88 percent which suggests inter label-correlations could aid in creating better models for APRT.