Hotel Review Classification Based on the Text Pretraining Heterogeneous Graph Neural Network Model.

Area of Science:

• Artificial Intelligence
• Natural Language Processing
• Recommender Systems

Background:

• The growing volume of online information necessitates precise product recommendations based on user preferences.
• Current recommendation systems often rely on basic sentiment analysis of reviews, neglecting deeper user demands and reducing classification effectiveness.
• Effective utilization of user-generated content, like reviews, is crucial for enhancing recommendation accuracy.

Purpose of the Study:

• To develop and evaluate a novel model that integrates advanced natural language processing and graph neural networks for improved product recommendation.
• To address the limitations of traditional sentiment analysis in recommendation tasks by incorporating user preference mining.
• To enhance the accuracy of travel-related product classification by understanding nuanced user demands from reviews.

Main Methods:

• Fine-tuning Bidirectional Encoder Representation from Transformers (BERT) on a large dataset of 1.4 million hotel reviews to capture trip-related word representations.
• Employing a similarity fussy-matching method to identify the main topics within user reviews.
• Constructing a heterogeneous graph attention network (HGAN) with an attention mechanism to mine user travel preferences.
• Integrating BERT and HGAN to perform user preference-based classification for travel recommendations.

Main Results:

• The proposed model, combining BERT and HGAN, achieved an accuracy of 70% in travel type classification.
• Experimental results demonstrated that the new model significantly outperformed five other baseline models in the classification task.
• The fine-tuned BERT model effectively generated rich representations of trip-related terms, aiding in preference mining.

Conclusions:

• The integration of BERT and HGAN offers a superior approach to recommendation systems compared to traditional methods.
• The model's ability to mine user preferences from reviews enhances the effectiveness of product classification and recommendation.
• This approach provides a promising direction for developing more sophisticated and accurate recommender systems in e-commerce.