Multitask feature learning approach for knowledge graph enhanced recommendations with RippleNet.

Area of Science:

• Artificial Intelligence
• Information Retrieval
• Data Science

Background:

• Traditional recommender systems suffer from data sparsity and cold start problems.
• Knowledge graphs offer auxiliary information to improve recommender systems.
• Existing knowledge graph embedding methods in deep learning have limitations in fully extracting graph information.

Purpose of the Study:

• To propose the Ripp-MKR model, a multitask feature learning approach for knowledge graph-enhanced recommendations.
• To combine joint and alternating learning strategies for knowledge graphs and recommender systems.
• To leverage RippleNet concepts for improved user characteristic representation.

Main Methods:

• Developed Ripp-MKR, a deep end-to-end framework integrating knowledge graph embedding with recommendation tasks.
• Utilized cross and compress units for automatic latent feature sharing and learning high-order interactions.
• Incorporated user historical interaction data with knowledge graph information for user representation.

Main Results:

• Ripp-MKR demonstrated substantial performance gains over state-of-the-art baseline models.
• The model achieved significant improvements in movie, book, and music recommendation tasks.
• Extensive experiments on real-world datasets validated the model's effectiveness.

Conclusions:

• Ripp-MKR effectively enhances recommender systems by integrating knowledge graphs.
• The proposed multitask learning framework successfully addresses sparsity and cold start challenges.
• The model offers a promising direction for knowledge graph-enhanced recommendation systems.