Published Paper Details:

Online hotel reviews provide valuable insights into customer experiences; however, such reviews are often short and informal, making automated sentiment analysis challenging. Single classification models frequently struggle to capture sentiment accurately due to limited contextual information and lexical variability. To address this issue, this paper presents a comparative study of voting-based and stacking-based ensemble learning methods for sentiment analysis of hotel reviews. Reviews are preprocessed and decomposed into bigram phrases to analyze the sentimentsin hotel reviews. Multiple feature representations including TF-IDF, Word2Vec, and BERT embeddings are employed to capture lexical, semantic, and contextual information. In the voting ensemble, independent classifiers aggregate predictions using soft voting, while the stacking ensemble integrates base learner outputs through a meta-classifier. Experimental results demonstrate that stacking ensembles achieve more consistent performance on ambiguous hotel review phrases, whereas voting ensembles provide a computationally efficient and reliable baseline. The findings highlight the effectiveness of ensemble strategies for short-text sentiment analysis in hospitality review systems.