Health recommendation system based on sentiment analysis from social media

Abstract

    This research develops a personalized mental health recommendation system based on emotion analysis from social media. The main goal is to leverage textual data and social relationships to deeply understand users’ emotional states, thereby providing suitable recommendations to effectively support mental well-being. GraphMoodRF integrates three main components: Graph Convolutional Network (GCN) to compute the Sentiment Analysis Score (SA Score) from both post content and social network structure; Fuzzy Logic to classify the SA Score into emotional levels such as very negative to very positive; and Neural Collaborative Filtering (NCF) to learn nonlinear relationships between users and recommended actions. Experimental results on 1,000 tweets collected from Twitter show that the system achieves an F1-score of 0.87 in emotion classification and Precision@3 of 0.85 in recommendation tasks, outperforming models like LSTM, BERT, and traditional Collaborative Filtering. The main contribution of this research is to propose a new, effective, and highly scalable model that combines GCN, fuzzy logic, and NCF to personalize recommendations based on emotions. GraphMoodRec has strong potential for real-world applications in platforms that support real-time mental healthcare applications, thus addressing the growing demand for digital mental health interventions.