A GPT-Based Approach for Sentiment Analysis and Bakery Rating Prediction

This paper presents a comprehensive approach to predicting the ratings of bakery establishments on diverse online platforms using natural language reviews. The study incorporates Large Language Models (LLMs) and Aspect-Based sentiment analysis to discern nuanced sentiment scores within specific categories. Utilizing advanced machine learning and regression methods, the paper introduces a robust predictive framework that employs LLMs, exemplified by GPT-3.5 Turbo, to accurately infer sentiment from natural language phrases. Framework’s effectiveness is demonstrated in accurately forecasting bakery scores based on such sentiment analysis, obtaining a MAE of 0.27, demonstrating low error rates in comparison to other state-of-the-art models. The study also addresses challenges associated with sentiment analysis, including emojis and sarcasm, and underscores LLMs enhanced proficiency in handling intricate linguistic nuances. Taken together, this research highlights the potential of LLMs in constructing a precise predictive model for online platform ratings, offering valuable insights into consumer perceptions.