Leveraging BERT with Extractive Summarization for Depression Detection on Social Media

Abstract

In our current time, the well-being of a person is not only determined by the physical health, but also by their mental health. A lot of focus and effort have been spent into raising the awareness of this issue. One such effort comes from the field of computer science by utilizing data from social media to provide additional information in detecting these mental disorders. In this research, authors proposed the Bidirectional Encoder Representations from Transformers (BERT) with extractive summarization to preprocess data obtained from popular social media platform such as Reddit and Twitter. The proposed method evaluated by compare it to several other methods in text classification to Figure out which method is best classifying whether someone is depressed or not. This research also contributes to evaluate whether applying extractive summarization to a document prior to feeding the data into BERT will contribute to an increase in performance when compared to the current method of data preprocessing, by measuring the accuracy, precision, recall and F1-score of each model. The evaluation result found that our proposed method by applying extractive summarization to reduce the sequence length achieved a better result compared to BERT base model.