Accurate multi-behavior sequence-aware recommendation via graph convolution networks.

Area of Science:

• Computer Science
• Artificial Intelligence
• Machine Learning

Background:

• Multi-behavior recommender systems utilize diverse user actions to improve recommendation accuracy.
• Existing methods often neglect the individual impact of specific behaviors on user preferences.
• Personalized recommendations enhance user experience across e-commerce, streaming, and content platforms.

Purpose of the Study:

• To propose an accurate framework for multi-behavior recommendations that captures both behavioral dependencies and individual behavior importance.
• To enhance recommendation performance by learning nuanced user preferences from sequential behavior data.

Main Methods:

• Developed MBA (Multi-Behavior sequence-Aware recommendation via graph convolution networks).
• Learned embeddings reflecting dependencies and relative importance of user behaviors.
• Employed sophisticated sampling strategies considering the sequential nature of behaviors during training.

Main Results:

• MBA demonstrated superior performance compared to existing multi-behavior recommendation methods.
• Achieved significant improvements in Hit Rate@10 (HR@10) by 11.2% and Normalized Discounted Cumulative Gain@10 (nDCG@10) by 11.4% on real-world datasets.
• Validated the effectiveness of sequence-aware learning and behavior importance in recommendation.

Conclusions:

• MBA provides accurate and personalized recommendations by effectively modeling user behavior sequences and their importance.
• The proposed framework advances multi-behavior recommendation systems, offering better user engagement and satisfaction.
• Highlights the significance of incorporating sequential information and behavior weighting in recommender models.