Understanding reviews at different levels has become essential when tracking customers and users feedback for different services and products. These reviews can be understood at a higher level if the aspect terms are grouped into coherent clusters. This paper presents a multi-task learning approach to Aspect-Based Sentiment Analysis (ABSA) that jointly performs aspect term extraction, aspect level sentiment classification, and cluster level prediction by leveraging unsupervised clustering of aspect terms to enrich the semantic representation and improve overall prediction consistency. Using the SemEval-2014 Task 4 restaurant reviews dataset, we apply K-means clustering on contextual embeddings of the aspect terms. These cluster assignments are then used as additional supervision in our model. The proposed model produces an accuracy of 83%. The cluster-based approach also provides more interpretable results by revealing broader thematic categories in customer feedback.