Improving Aspect-Based Sentiment Analysis for Hotel Reviews with Latent Dirichlet Allocation and Machine Learning Algorithms

Abstract

The rapid expansion of online platforms has resulted in a deluge of user-generated content, emphasizing the need for sentiment analysis to gauge public opinion. Aspect-based sentiment analysis is now essential for uncovering intricate opinions within product reviews, social media posts, and online texts. Despite their potential, the complexity of human emotions and diverse language nuances pose significant challenges. Our study focuses on the importance and trends of sentiment and aspect-based sentiment analysis in automated review analysis, with a primary focus on Indonesian-language hotel reviews. Our research underscores the need for nuanced tools to unravel multifaceted sentiments. We propose an automation framework that utilizes Latent Dirichlet Allocation (LDA) for feature extraction. We evaluate LDA's performance, enhance it through filtration, and enrich it by integrating it with Word2Vec and Doc2Vec. Our methodology encompasses various machine learning algorithms, including Logistic Regression (LR), Stochastic Gradient Descent (SGD), Support Vector Machine (SVM), Random Forest (RF), and Light Gradient Boosting Machine (LGBM). Empirical results reveal that the optimal combination involves LDA bigram and Word2Vec, alongside the LGBM classifier, yielding an average F1 score of 86.6 across ten aspects. This contribution advances automated aspect-based sentiment analysis, offering concrete implications for e-commerce, marketing, and customer service. Our insights inform precise marketing strategies and enhance customer experiences, underscoring the research's relevance in the digital landscape.