An Improved Dual-Channel Deep Q-Network Model for Tourism Recommendation.

Area of Science:

• Artificial Intelligence
• Data Science
• Tourism Informatics

Background:

• Traditional tourism recommendation systems suffer from low accuracy and personalization due to sparse data.
• Implicit features like context, travel trajectories, and comments are underutilized in existing models.

Purpose of the Study:

• To develop an advanced tourism recommendation system leveraging deep learning.
• To enhance recommendation accuracy and user personalization by incorporating diverse data sources.

Main Methods:

• Utilized Long Short-Term Memory (LSTM) networks for feature extraction from contextual information, scenic spot data, and tourist comments.
• Implemented a Deep Q-Network (DQN) with a dual-channel mechanism, analyzing user online behavior and long-term preferences via feedback loops.
• Developed a recommendation strategy with value evaluation and target networks for optimal strategy learning.

Main Results:

• The proposed model demonstrated an average accuracy increase of 76.61% compared to baseline models.
• Achieved an average increase of 43.48% in normalized discounted cumulative gain over baseline methods.
• Successfully trained and validated on Yelp, DP, and Tourism datasets across multiple scenarios.

Conclusions:

• The novel LSTM-DQN approach significantly improves tourism recommendation services.
• The model effectively addresses limitations of traditional methods by utilizing implicit features for enhanced personalization and accuracy.
• This research offers a robust framework for personalized tourism recommendations in diverse settings.