DEPRESSION DETECTION BY ANALYZING SOCIAL MEDIA POST OF USER: A REVIEW
Prof. Dhanashree Patil, Yogesh Palve, Ketaki Todkar, Akash Shinde, Anisha Biradar, Tejaswini Patil
Abstract— In contemporary times, identifying depression at an early stage has become a critical issue within the field of psychology. Mental health problems, notably depression, afflict a significant portion of the global population, with an estimated 300 million individuals currently affected. Leveraging the vast amount of user-generated content on social media platforms, researchers are exploring ways to utilize this data to identify potential mental health issues in users. Depression remains a pressing concern in society and a topic of ongoing research globally. Despite numerous studies attempting to understand various mood disorders such as depression, anxiety, and stress using activity logs from devices like smartphones, accurately predicting depressive episodes remains challenging. Social media analysis has emerged as a popular approach to tackle this challenge. In this paper, we propose a system for assessing depression and detecting suicidal ideation by predicting suicidal tendencies based on the severity of depression. Our study aims to develop effective machine learning algorithms for identifying potential signs of depression in individuals through their social media posts. To achieve this, we trained and tested classifiers to discern whether a person exhibits signs of depression using features extracted from their social media activities. Various machine learning algorithms were employed to train and classify users into different levels of depression on a scale ranging from 0 to 100%. Data in the form of posts were collected and categorized into depressed or non-depressed using machine learning algorithms. Our primary contribution lies in exploring a range of features and their effectiveness in detecting the degree of depression. We aim to build a deep learning model capable of categorizing users with depression by learning from individual-level labels to infer post-level labels. By amalgamating the probabilities of post label categories, we can create temporal posting profiles to classify users with depression. Our findings reveal distinct variations in posting behaviors between depressed and non-depressed users, as indicated by the combined probabilities of post label categories. In our study, we utilized natural language processing (NLP) techniques, employing the BERT algorithm, to efficiently identify potential depression indicators in social media content. This approach offers a more accessible and effective means of detecting depression.
Keywords— Machine Learning, NLP, BERT Algorithm, Depression, Classification, Social Media Post.
