Aspect-Based Sentiment Analysis of Google Maps Reviews For Barbershop Competitor and Brand Strategy Analysis Using Machine Learning

Abstract

Customer reviews on Google Maps contain information regarding service experiences but are unstructured and may discuss multiple aspects within a single review. This study applies Aspect-Based Sentiment Analysis (ABSA) to analyze customer perceptions across 23 barbershops and compare the performance of Naive Bayes, Support Vector Machine (SVM), Random Forest, and IndoBERT. A total of 5,418 reviews were collected from the selected barbershops and processed through cleaning, preprocessing, aspect identification, and aspect-level lexicon-based sentiment labeling. After cleaning and deduplication, 5,299 reviews generated 7,457 aspect-level observations classified into positive, neutral, and negative sentiment. Naive Bayes, SVM, and Random Forest used TF-IDF representations, while IndoBERT used contextual transformer-based representations. Evaluation using accuracy, precision, recall, Macro F1-score, and ROC-AUC showed that IndoBERT achieved an accuracy of 98.57%, precision of 93.28%, recall of 93.99%, Macro F1-score of 93.55%, and ROC-AUC of 99.88%. General, Result Quality, and Service were the most frequently identified aspects, while Cleanliness, Comfort, and Service had the highest proportions of positive sentiment. Waiting Time and Price showed relatively higher proportions of negative sentiment. A Wilson Score Lower Bound was used as supplementary descriptive information for comparing the relative positions of the 23 barbershops. Because sentiment labels were automatically generated using a lexicon-based procedure, the classification and competitor-analysis results represent performance and descriptive patterns based on weak labels rather than manually annotated ground truth.