Abstract:
"This study aims to analyse Sri Lankan restaurant reviews to pinpoint key characteristics highlighted by
users. By extracting these features, sentiment scores can be assigned to evaluate user preferences for
specific aspects of the product or service. While many recommendation systems rely on English
datasets, this research focuses on enhancing recommendation accuracy by incorporating insights from
English reviews. Extracting features from English restaurant reviews and providing polarity scores for
both the overall product/service and the specific features identified enables users to gain a
comprehensive understanding of the offerings, thus improving recommendation accuracy.
The proposed methodology uses SVM (Support Vector Machine) and GBR (Gradient Boosting
Regressor) ensemble methods to determine review polarity through contextual analysis. This study will
explore how developing an ensemble approach can enhance the overall accuracy of sentiment
classification. A dataset of restaurant reviews from TripAdvisor and Kaggle is used to assess the
method. The results demonstrate that these models function effectively in NLP applications, advancing
the field of Aspect-Based Sentiment Analysis, with trials showing the proposed methodology achieving
a high accuracy rate of 93% in correctly identifying the sentiments of restaurant reviews."
