In the ever-evolving landscape of the banking industry, the invaluable role of customer feedback cannot be overstated. In an era dominated by digital communication, customer reviews and textual comments wield a considerable influ-ence over consumer choices, shaping the overall reputation and performance of financial institutions. The multifaceted nature of language and the sheer volume of textual data, however, present a formidable challenge when attempting to extract meaningful insights from these reviews. This research embarks on a pioneering journey to navigate this challenge by adopting a sophisticated approach to sen-timent analysis. The traditional methods of sentiment analysis fall short when faced with the complexity and nu-ance inherent in customer reviews. To overcome these limi-tations, we embrace cutting-edge technologies, specifically Bidirectional Encoder Representations from Transformers (BERT) with PyTorch. The central objective of this research is to construct an ad-vanced sentiment analysis system that goes beyond mere classification of customer reviews as positive or negative. We delve into the nuances of customer sentiments and preferences, aiming to unravel the intricate layers of feed-back intricately woven into the vast tapestry of textual data. This ambitious project focuses on a substantial dataset comprising 10,000 recent customer reviews sourced from 48 diverse banks across the United States. The integration of BERT embeddings with PyTorch signifies a departure from conventional sentiment analysis methods. BERT, known for its contextual understanding of language, enhances the model's ability to capture subtleties within the reviews, surpassing the limitations of traditional methods. The primary goal is not merely to categorize reviews but to provide the banking industry with nu-anced and precise insights into customer sentiments. By leveraging the power of BERT embeddings with PyTorch, we aim to overcome the challenges inherent in textual feedback analysis and elevate the analysis to a level where it becomes a strategic tool for enhancing service quality. This research aspires to empower the banking industry, offering actionable insights that transcend the binary classifi-cation of positive or negative sentiments. The vision ex-tends beyond classification; it encompasses a comprehensive understanding of the underlying sentiments and preferences expressed by customers. Ultimately, this endeavor seeks to catalyze a paradigm shift in how the banking industry interprets and responds to customer feedback, fostering an environment of continuous improvement and heightened customer satisfaction.
The dataset utilized in this research comprises over 10,000 customer reviews sourced from 48 distinct banking institutions within the United States. This publicly available dataset can be accessed through the Kaggle community platform, offering a comprehensive collection of customer sentiments towards various banks. The dataset's primary objective is to serve as a resource for training and evaluating the sentiment analysis system proposed in this research. For easy accessibility, the dataset can be downloaded using the following link: https://www.kaggle.com/datasets/trainingdatapro/20000-customers-reviews-on-banks/data. The richness of this dataset not only allows for a detailed exploration of customer feedback but also ensures a diverse and representative sample, crucial for the effectiveness and generalizability of the sentiment analysis model.
The purpose of the project is to build the sentiment analysis using BERT to analyze each customer review to judge that it is positive or negative review, considering 4-5 stars as positive reviews and 1-3 stars as negative reviews.