Automatic personality prediction from Indonesian user on twitter using word embedding and neural networks

Abstract

Personality can be shown from user’s generated content and their activities. In this research, we attempt to predict Twitter user’s personality strictly from their tweets based on Big Five. We limit the user to Indonesian only, thus also limit the diversity of language. We construct word embedding as the input for our neural network, which uses LSTM, Bi-LSTM, and GRU. Using F-Measure, we obtained 0.82812 for highest averaged training score when vectors are averaged, using GRU with ReLU as activation function. We also observe the consistency of high result on all but low result on conscientiousness, even though the imbalance of dataset also occurs on extraversion.