Unsupervised Sentiment Analysis using Hybrid Clustering and Word Embeddings for Enhanced Sentiment Polarity Detection

Unsupervised sentimental analysis is an effective approach for extracting opinions and emotions from text data without relying on labeled datasets. This method leverages techniques such as clustering, topic modeling, word embeddings, and lexicon-based analysis to identify sentiment patterns in unstructured text. Commonly used techniques include Latent Dirichlet Allocation (LDA) for topic modeling, word vectorization methods like Word2Vec and GloVe to capture semantic relationships, and lexicon-based approaches that utilize predefined sentiment dictionaries. Additionally, deep learning models such as autoencoders and GAN are explored for feature extraction and representation learning in sentiment analysis. Our proposed method integrates word embeddings with a hybrid clustering approach combining K-Means and Hierarchical clustering to improve sentiment polarity detection in large-scale datasets. By leveraging the strengths of both density-based and centroid-based clustering, our method ensures more accurate grouping of similar sentiments without prior annotations. The results obtained on the taken datasets demonstrate that the proposed approach enhances sentiment classification accuracy compared to the traditional methodologies. This research contributes to the advancement of sentiment analysis by offering an effective, scalable, and domain-independent methodology for sentiment detection in various applications, including social forum monitoring, product reviews, and market analysis.