Deep Rating and Review Neural Network for Item Recommendation

Area of Science:

• Artificial Intelligence
• Machine Learning
• Natural Language Processing

Background:

• Recommender systems often struggle with data sparsity.
• Existing methods use review text as auxiliary data but rely solely on ratings for training.
• This reliance on ratings can lead to loss of semantic information present in reviews.

Purpose of the Study:

• To propose a novel deep model, the deep rating and review neural network (DRRNN), for recommendation.
• To address the limitations of existing models by incorporating both review text and ratings more effectively.
• To enhance recommendation quality by preserving richer semantic information from user reviews.

Main Methods:

• Developed a deep rating and review neural network (DRRNN).
• DRRNN utilizes both target rating and target review as ground truth for error backpropagation during training.
• This contrasts with traditional methods that only use ratings.

Main Results:

• The proposed DRRNN model demonstrates effectiveness in rating prediction.
• Extensive experiments on four public datasets validate the model's performance.
• DRRNN successfully retains more semantic information from reviews compared to existing approaches.

Conclusions:

• DRRNN offers an improved approach to recommender systems by leveraging both ratings and review text.
• The model effectively mitigates information loss inherent in rating-only training methods.
• DRRNN enhances the quality of recommendations by better utilizing the richness of review data.