Leveraging Large Language Models for Aspect-Based Sentiment Analysis: A Restaurant Recommendation System for Entrepreneurs in Lisbon
TL;DR
The main conclusion is that LLMs provide a robust tool for interpreting large volumes of textual review data, enabling the creation of a prototype decision-support tool that offers entrepreneurs detailed competitive insights.
Abstract
Abstract The purpose of this study is to implement a restaurant recommendation system tailored for entrepreneurs within the Lisbon Metropolitan Area (AML), utilising Large Language Models (LLMs) for aspect-based sentiment analysis (ABSA) of online customer reviews. The methodology involved three phases: first, pre-processing restaurant and review data sourced from the DIG-IN online platform; second, defining relevant customer satisfaction attributes based on adapted service quality models and extracting associated keywords using LLMs; third, developing and applying prompts to GPT-4, GPT-3.5 Turbo, and Mistral 7B Instruct models to classify sentiment polarity across defined attributes like food quality, service, and ambiance found within 4,464 reviews from 115 ‘Italian’ restaurants in the AML. Concordance tests confirmed the models’ precision against human evaluation, with GPT-4o achieving 81.6% agreement. Results demonstrated the viability of using LLMs to extract, identify, and classify relevant attributes for sentiment analysis. The main conclusion is that LLMs provide a robust tool for interpreting large volumes of textual review data, enabling the creation of a prototype decision-support tool that offers entrepreneurs detailed competitive insights. This research demonstrates the practical application of LLMs and ABSA for strategic investment decisions in the highly competitive restaurant sector.