Role of Artificial Intelligence in Sentiment Analysis as State-of-the-Art in Future Business Scenarios

Abstract

Any organization, whether it is focused on business or products, must design scalable and efficient sentiment analysis tools in light of the increasing reliance on digital data. Businesses of all sizes must analyze sentiment in order to understand and quickly address customer opinions and comments on a variety of subjects. Businesses depend on feedback to understand customer emotions precisely after improve their processes within marketing and customer service as well as product development. Sentiment analysis using artificial intelligence, namely machine learning, deep learning, and Natural Language Processing (NLP), is essential for text sentiment categorization as positive, negative, or neutral. This study examines the use of Artificial Intelligence specifically in sentiment analysis (SA) with emphasis on Distil BERT and AlBERT, advanced Natural Language Processing models on Flipkart product reviews sentiment classification. The raw data, containing over 200 thousand records of customers’ feedback, has fields such as product name, price, rating, review text, summary, and sentiment. Despite its simple and intuitive nature, the preparation of the given data involves various preprocessing steps like normalization, lemmatization, stop word removal, and data balancing through resampling methods. Both Distil BERT and Albert are trained and tested by using certain benchmarks (accuracy, precision, recall, and F1score) for measuring their performance. Distil BERT had the highest performance with an accuracy of 94.90%, precision 94.91%, recall 94.90%, and F1score of 94.90%, followed by Albert with an accuracy of 93.33% across the evaluation metrics. These models were also characterized by good percentage linger around the ROC curves and nice error matrix, which confirmed the good classification capability between positive, neutral, and negative sentiments. In contrast, traditional ML models like Random Forest and SVM yielded significantly lower accuracy scores of 75.52% and 75.80%, respectively. The research demonstrates how transformer-based models may be used in future business situations for efficient sentiment analysis and product review assessment, allowing companies to use consumer input to improve customer satisfaction and decision-making