M²: Mixed Models With Preferences, Popularities and Transitions for Next-Basket Recommendation.

Area of Science:

• Computer Science
• Artificial Intelligence
• Machine Learning

Background:

• Next-basket recommendation aims to predict a set of items users will purchase together.
• Existing methods often rely on complex recurrent neural networks.

Purpose of the Study:

• To develop a novel mixed model (M²) for next-basket recommendation.
• To effectively model user preferences, item popularities, and item transition patterns.

Main Methods:

• Developed a mixed model (M²) incorporating preferences, popularities, and transitions.
• Utilized a simple encoder-decoder approach (ed-Trans) for modeling item transitions.
• Compared M² against five state-of-the-art methods on four benchmark datasets.

Main Results:

• M² significantly outperformed all tested state-of-the-art methods across all tasks (first, second, and third next basket).
• Achieved performance improvements of up to 22.1% compared to existing approaches.
• The ed-Trans method proved more effective than recurrent neural networks for recommendation performance.

Conclusions:

• The proposed M² model offers a superior approach to next-basket recommendation.
• The ed-Trans component is a key factor in the model's enhanced performance.
• The study provides valuable insights into experimental protocols and evaluation metrics for this task.