Work

Graph-Enhanced sentiment Analysis for Hotel Reviews using neural networks

Year: 2025
Type: dissertation
Abstract: Sentiment analysis of hotel reviews is important for figuring out how happy customers are and making better business decisions. But traditional models only look at reviews as separate documents and don't consider the natural relationships between reviews that have similar contexts or feelings. This study introduces a novel graph-enhanced sentiment analysis framework that integrates Graph Neural Networks with contextual word embeddings, representing reviews as semantically interconnected nodes within a graph. A lot of tests on 12,000 TripAdvisor reviews show that the suggested framework is better than all the other ones that were tested. The best sentiment classification accuracy is 80.33%, which is 7.94% higher than the best traditional baseline accuracy of 74.42% on the same datasets. Ablation studies comparing architecturally similar GNNs and Multi-Layer Perceptrons demonstrate that the sole factor enhancing classification accuracy by 14.71 percentage points is the graph's structure. The framework also performs exceptionally well in predicting ratings; its RMSE is 0.8406, while the best baseline's is 0.9817. This study demonstrates that the integration of relational structure via graph neural networks yields significant and pragmatic enhancements in sentiment analysis within the hospitality sector, providing a novel framework for utilising document relationships in natural language processing projects.
Language: English
Open Access status: green
OpenAlex ID: W7204680987
Cites: 14
Cited by:
Related to: 10