Aspect-based sentiment analysis (ABSA) presents a notable challenge in the realm of sentiment analysis (SA) by emphasizing the identification of fine-grained opinion polarity directed toward particular aspects. This study introduces EXB_RNN, a novel ensemble learning framework for ABSA, adept at discerning nuanced opinion polarities toward specific aspects in texts. Integrating the strengths of ELMo, XLNet, and BERT for profound contextual word representations, bidirectional dependencies, and nuanced contextual embeddings, respectively, EXB_RNN leverages these transformer models as feature extractors for an RNN classifier. This classifier determines sentiment polarity for each aspect, employing a majority voting mechanism to enhance prediction accuracy. Demonstrating its effectiveness, EXB_RNN achieves outstanding accuracy metrics of 97.74%, 98.87%, 97.94%, and 99.05% across various Semeval and Twitter datasets, highlighting its robustness and precision in capturing aspect-level sentiments. This ensemble approach marks a significant advancement in ABSA, combining deep learning techniques to address the complexity of fine-grained sentiment analysis.