Measuring the Effectiveness of LDA-Based Clustering for Social Media Data

Abstract

Social media has come out as a great platform for users to communicate and share their opinions, photos, and videos that contemplate their moods, feelings, and emotions. This wide variety of data provides multiple possibilities for exploring social media data to investigate feelings and sentiments based on their moods and attitudes. With the enormous increase in mental health disorders among individuals, there is a massive loss in productivity and quality of life. Social media platforms like Reddit are used to seek medical advice on mental health issues. The structure and the content on various subreddits can be employed to interpret and connect the posts for mental health diagnostic problems. In this work, we have focused on seven mental health disorders, namely Anxiety, Depression, Bipolar, Autism, Borderline personality disorder, Schizophrenia, and mental health, which are actually subreddits posted by users on the Reddit social media platform. In this work, we have measured the effectiveness of topic modeling using Latent Dirichlet Allocation on these social media posts to identify the most used words and discover the hidden topics in their posts and also analyzed the results on evaluation metrics based on perplexity and coherence scores on unigrams, bigrams, and trigrams.