Personalized diversification of complementary recommendations with user preference in online grocery

Area of Science:

• Computer Science
• Artificial Intelligence
• E-commerce Technology

Background:

• Complementary recommendations are crucial for cross-selling, but customer shopping behaviors vary significantly.
• Existing recommendation systems struggle to cater to both exploratory (diverse) and conventional (focused) shopping intents.
• Current diversification methods often increase heterogeneity, failing to address the need for homogeneity in conventional complementary shopping.

Purpose of the Study:

• To develop personalized diversification strategies for complementary recommendations that adapt to user shopping intent.
• To address the limitations of existing methods in modeling both heterogenized and homogenized recommendation preferences.
• To improve the adaptability of recommendation systems based on distinct user shopping behaviors.

Main Methods:

• Proposed two diversification strategies: heterogenization and homogenization, utilizing determinantal point processes (DPP).
• Estimated user intent for exploratory or conventional shopping based on transaction history.
• Developed a dynamic algorithm to personalize diversification strategies using estimated user intent scores.

Main Results:

• Demonstrated the effectiveness of the proposed re-ranking algorithm on the Instacart dataset.
• Successfully modeled and addressed both heterogenized and homogenized complementary recommendation needs.
• Showcased the ability to dynamically personalize diversification strategies.

Conclusions:

• Personalized diversification strategies can effectively cater to diverse user shopping intents in complementary recommendations.
• The proposed DPP-based approach offers a novel solution for balancing recommendation diversity and homogeneity.
• This research advances the field of personalized e-commerce by adapting recommendations to specific user behaviors.