Knowledge-reinforced explainable next basket recommendation.

Area of Science:

• Artificial Intelligence
• Information Retrieval
• Recommender Systems

Background:

• Next basket recommendation models user purchasing behavior through basket sequences.
• Current methods often act as black boxes, neglecting explainability and user satisfaction.
• Existing approaches primarily target consumer users, overlooking business user specificities.

Purpose of the Study:

• To develop an explainable next basket recommendation system tailored for business users.
• To integrate knowledge graphs and reinforcement learning for improved recommendation accuracy and interpretability.
• To address the limitations of existing black-box models in providing user-centric explanations.

Main Methods:

• Constructed a basket-based knowledge graph to derive rich entity embeddings.
• Fused user embeddings, basket sequence embeddings, and repurchase embeddings for predictive vectors.
• Employed reinforcement learning for path reasoning to generate explainable recommendations.

Main Results:

• The proposed Knowledge Reinforced Explainable Next Basket Recommendation (KRE-NBR) system demonstrated superior performance over state-of-the-art baselines.
• Successfully generated explanations for next basket recommendations, a novel contribution.
• Effectively modeled business user repurchase behavior through specialized embeddings.

Conclusions:

• KRE-NBR offers a significant advancement in explainable next basket recommendation, particularly for business contexts.
• The integration of knowledge graphs and reinforcement learning enhances both recommendation quality and transparency.
• This work pioneers explainable recommendations in the next basket prediction domain, improving user trust and satisfaction.