Aspect-Based Sentiment Analysis (ABSA) is a critical task in natural language processing (NLP) that aims to identify and analyse sentiments towards specific aspects in textual data. This paper proposes an ensemble learning model for ABSA that combines the strengths of three state-of-the-art transformer-based models, namely XLNet, XLM, and BERT, with a Support Vector Machine (SVM) classifier. The proposed hybrid ensemble model seeks to leverage the unique features of XLNet, XLM, and BERT to enhance ABSA performance by capturing fine-grained sentiment information and contextual dependencies. XLNet utilizes an autoregressive framework to capture bidirectional dependencies, XLM specializes in cross-lingual language modeling, and BERT employs a masked language modeling approach for contextualized representations. By combining these models, we aim to exploit diverse contextual information and improve the overall sentiment analysis accuracy at the aspect level. In our approach, the outputs of XLNet, XLM, and BERT are used as input to an SVM classifier separately. The SVM classifier is trained to classify the sentiment polarity of aspects based on the features extracted and the ensemble classifier used for final predictions. The proposed hybrid ensemble model (BX2_Ensemble) with XLNet, XLM, and BERT, along with the SVM classifier, offers a promising approach for ABSA applications, providing enhanced performance and robustness in capturing sentiment information at the aspect level with improved sentiment analysis accuracy of 98.01%, 97.87% and 97.74% on three different datasets which are based on Amazon reviews, Twitter and TripAdvisor datasets.

Ensemble Learning, XLNet, XLM, BERT, ABSA, SVM, Hybrid Model, Transformer models.