Enhancing Minority Sentiment Classification in Gastronomy Tourism: A Hybrid Sentiment Analysis Framework with Data Augmentation, Feature Engineering and Business Intelligence

Abstract

The gastronomy tourism industry plays an important role in boosting local economies, enhancing the travel experience, and preserving culinary traditions unique to specific places. In this context, comprehending customer sentiments is of paramount importance for business decision-making, menu choice offerings, marketing strategies, and customer service improvements. Traditional sentiment analysis methods in gastronomy tourism tend to be time-consuming, prone to human error, and influenced by subjectivity. Furthermore, the absence of an effective visualization strategy hampers the reliability of sentiment analysis efforts. Compounding this, the data collected also often lacked balance across sentiment classes, making it challenging to predict minority sentiments accurately. To address these challenges, our research introduces a hybrid approach, combining various lexicon-based sentiment and emotional analysis algorithms, thereby enhancing the reliability of customer review analysis in the gastronomy tourism sector. Subsequently, we optimize machine learning sentiment classification by employing data augmentation in conjunction with feature engineering strategies, to improve the recognition of minority sentiment classes. Additionally, we present a comprehensive business intelligence and visualization solution that is personalized for the gastronomy tourism industry in Sarawak and offers real-time sentiment visualization. The optimization of sentiment classification, achieved through the integration of synonym augmentation and n-gram feature engineering in conjunction with kNN classifiers, has yielded impressive results. This approach attains optimal classification performance, boasting an accuracy rate of 0.98, a F1-score and a ROC-AUC score of 0.99. Notably, this methodology significantly enhances the recognition of minority sentiment classes within the dataset, addressing the main challenges in this research.