Analysis of the Performance of Transformers and Ensemble Schemes for Twitter US Airlines Sentiment Analysis

Undoubtedly, sentiment analysis is an important part of NLP, basically helping machines to capture human emotions; most of the collected datasets on sentiments are created to help a company understand the needs of its customers and work for improvement in services.When these datasets are made publicly available, customers can also benefit by learning the experiences of other clients with certain companies for informed decisions on products or services they may want to buy.Researchers use these datasets in training artificial intelligence models to better their accuracy and performance.Beyond the academic and research purposes, the commercial value is evident as well.The rapid growth of data makes it impossible for humans to manually categorize and assess whether customers are satisfied or not in a reasonable time.Therefore, automated sentiment analysis becomes essential.The purpose of this work is to propose a method that improve the performance of six different Transformer models.We will explore how the majority vote ensemble exploits the diversity of models and achieves an 5% increase.