Measuring Public Perception of Depression on Twitter: A Sentiment Analysis Study

Businesses, enterprises, and various organizations use social media as an essential data repository for analytics to gain commercial benefits and to understand the opinions of customers and the public in general. This study aims to measure the public opinions on depression and to help recognise individuals who may suffer from a depressive disorder. Labelled data about whether an individual has depression or not is sourced from Kaggle and unlabelled data is extracted from Twitter which is both then analysed in this paper. Various classification models have been implemented for classifying sentiments and detecting depression, including Multilayer Feedforward Neural Network, Convolutional Neural Network, Long Short Term Memory, Recurrent Neural Network, Gated Recurrent Unit, and Transformer. In addition, Latent Dirichlet Allocation (LDA), a topic modelling technique, is applied to identify important and essential words (indicators) related to depression. The data is first collected, then pre-processed, tokenized and padded sequentially, and finally analysed to detect depression by using various neural network algorithms. On the tokenized text LDA is applied. The entire analytical process is presented and discussed in detail. Word clouds are generated for both categories of users with some important words being highlighted. LDA helps us to find certain keywords uniquely associated with both categories. The findings of this study have shown some important implications and can be used by organizations and healthcare practitioners in treating depression. The analytical procedure provided considers many factors and therefore can potentially assist with the pre-screening process for diagnostical usage. In addition, the analytical models can be used by mental health professionals in treating depressed patients and monitoring a patient's progress in terms of medication prescription, therapy, etc.