A Sentiment Analysis Based on the Comparison of BERT and Naive Bayes
Dhruv Singh Bisht, Alka Pant, Jaishankar Bhatt · 2024
In recent years sentiment analysis has become a very important part of natural language processing. It allows us to automatically understand the sentiments of humans and views present in textual data. Due to its highly accurate information acquisition based on the context, Bidirectional Encoder Representations from Transformers (BERT) have become well-known in this field, among several different approaches. The main objective of this research paper is to discuss the modeling, procedure of training, and assessment of the performance of a sentiment analysis BERT model and compare its performance with a probability-based traditional Naive Bayes approach. We try to discover how we can fine-tune BERT for sentiment analysis tasks, and examine its several model configurations and different ways for pre-processing data. Moreover, we conduct detailed tests on well-known datasets to evaluate how accurately the model identifies sophisticated feelings in various environments. We try to show the efficacy of the constructed BERT model in accurately figuring out sentiment polarity and understanding the nuances of human-like sentences in text with the help of detailed studies and comparing ours with different present approaches. This research demonstrates how sentiment analysis models based on BERT can extend their applications in several domains by paving the way for natural language processing systems to understand sentiments easily and more accurately.