2020 Machine Learning class final project @ UT Austin ischool
Won the First Place from Popular Vote for Best Project Award
See class website
Using machine learning to predict personalities is not only beneficial for customizing users' experience of interacting with products, but could further improve their experience of using artificiall intelligence products.
This paper used natural language processing algorithms to predict MBTI personalities and extend the dataset scope to include Azure sentiment score and social media posts to testify prediction performance. It compared models of GRU, LSTM and RNN with different hyperparameters of node numbers, dropout regularization and epochs, and found the best prediction accuracy of 0.938.
Meanwhile, it measured the influence of Azure sentiment score on prediction accuracy and the effectiveness of algorithm in predicting Tweets. Although results find the latter two efforts futile, the prediction results might be improved in the future by expanding dataset and incorporating more features.