CFSH: Factorizing sequential and historical purchase data for basket recommendation.

Area of Science:

• E-commerce analytics
• Machine learning for retail

Background:

• Next-basket prediction is vital for e-commerce.
• Existing methods include item-centric (suffers from data sparseness) and user-centric (ignores sequential behavior) paradigms.

Purpose of the Study:

• To introduce a hybrid model, Co-Factorization over Sequential and Historical purchase data (CFSH), for enhanced next-basket recommendation.

Main Methods:

• Developed a hybrid approach, CFSH, combining item-centric and user-centric paradigms.
• Mined global sequential patterns to mitigate data sparseness.
• Simultaneously factorized product-product and customer-product matrices to learn representations.

Main Results:

• CFSH effectively addresses the data sparseness issue.
• The model leverages both sequential and historical purchase data for improved customer and product representations.
• Achieved superior performance in next-basket prediction compared to state-of-the-art methods on real-world datasets.

Conclusions:

• The proposed CFSH model offers a more robust and effective solution for next-basket recommendation.
• Hybrid approaches integrating sequential and historical data yield better prediction accuracy.
• CFSH demonstrates significant improvements over existing methods in predicting customer transactions.