| Harish Palakurthi |
The proliferation of online reviews has made consumer sentiment a critical data point for businesses of all sizes. However, discrepancies between overall star ratings and the textual content of reviews can complicate sentiment analysis tasks. This study investigates the effectiveness of incorporating aspect identification within reviews to improve sentiment prediction accuracy. We compare four distinct approaches: traditional machine learning models, advanced OpenAI large language models (LLMs), SAS Model Builder, and the Sentiment Intensity Analyzer (VADER). The evaluation employs a benchmark dataset of consumer reviews paired with corresponding star ratings and the sentiment extracted from the ratings. The sentiment of consumer reviews was analyzed using the four methods. The results demonstrate that OpenAI LLMs achieve superior sentiment prediction accuracy (88%) followed by aspect-based sentiment analysis using traditional machine learning models (82.5%), SAS Model Builder (75%), and the simple Sentiment Intensity Analyzer-VADER (66%). These findings highlight the value of aspect-based sentiment analysis for bridging the gap between ratings and reviews, keeping the computational requirements of methods used in mind. The aspects provide insights into pain points of consumers and strengths of products and services of businesses. The research offers valuable insights into the strengths and limitations of each method, informing potential strategies to mitigate discrepancies in sentiment perception. |