A Sentiment-Aware Hybrid Ensemble Clustering Framework for Social Network Content Analysis and Community Discovery

Abstract

Social media stages produce huge volumes of short, familiar, and sentiment-rich content that offerings significant encounter for community discovery and thematic analysis. Traditional clustering approach struggle with sparsity, noise, and limited surroundings information present in such texts. To address these limitations, this study introduces a sentiment-aware hybrid ensemble clustering framework designed openly for social network content analysis. The planned method integrates lexical features (TF-IDF), semantic illustration derived from Sentence-BERT embeddings, and sentiment division scores to imprisonment both topical and affecting scopes of user-generated posts. Three fundamentally different clustering algorithms K-Means, Agglomerative Clustering, and DBSCA are collective through a consensus fusion implement that enhances constancy, reduces algorithmic bias, and progresses strength in noisy environment. The framework is assessed on both synthetic Facebook-brand data and a real-world Kaggle Twitter dataset to assess generalizability. Results validate that the hybrid ensemble approach yields superior cluster steadiness, sentiment homogeneity, and interpretability compared to divide base models. Visual analyses using PCA estimates and group cluster mappings further validate the framework’s efficiency for revealing latent community and sentiment-driven behavioral patterns. This study highlights the value of multi-view clustering for social network mining and distributes an understandable, scalable solution for requests in digital marketing, community monitor, and customer engagement analytics.