Abstract:

Abstract Due to the increasing number of reviews, it has become more important for businesses to analyze their customer's sentiments. This paper presents a framework that uses pre-trained language models such as BERT, XLNet, and Electra to analyze these sentiments. The framework is based on the Sentiment140 dataset which contains over 1.6 million tweets with tags. This collection of sentiments allows us to perform an evaluation of the models' performance. The goal of this paper is to analyze the effectiveness of these models in categorizing and understanding the sentiments in customer reviews. BERT, for instance, has demonstrated exceptional performance in various tasks related to natural language processing. Another model that is transformer-based is XLNet, which adds more capabilities by utilizing permutation-based learning. On the other hand, the new generation of model, known as Electra, focuses on the generator discriminator learning. Through the incorporation of these models, we can leverage the contextual understanding of the sentiments in the customer reviews. In this paper, we thoroughly examine the performance of the different models in the framework for sentiment analysis. We tested their precision, recall, F1-score, and accuracy in identifying and categorizing the sentiments in customer reviews. We also discuss the impact of adjusting the models on the task, as well as the tradeoffs between performance gains and computational resources. The findings of the study provided valuable information on the utilization of pre-trained models for analyzing customer reviews. We analyzed the performance of the different models BERT, XLNet, Electra, and BERT, revealing their weaknesses and strengths. This helps businesses identify the best model for their sentiment analysis needs. The study's findings have contributed to the advancement of sentiment analysis and natural language processing. It offers valuable recommendations that will aid in the future research efforts.

Keywords: