Machine Learning and LLM Integration for Online Mental Support

Abstract

• Pioneered the development of a dynamic pipeline aimed at nurturing mutual support within online communities of individuals navigating similar mental health challenges, with a primary focus on the social media platform Reddit. • Employed feature engineering methodologies, leveraging Sentence Transformers based on BERT architecture, to extract embeddings from both user posts and responses. Utilized these embeddings to implement a KNN algorithm, facilitating the identification of highly relevant and supportive responses to users’ posts. • Seamlessly integrated Language Model (LLM) capabilities, specifically GPT-3, to augment the pipeline’s functionality. Leveraged the selected supportive responses as prompts for the LLM, generating empathetic and appropriate responses to users’ posts, fostering a supportive and empathetic online environment for mental health discourse.