Personalized bundle recommendation using preference elicitation and the Choquet integral.

Area of Science:

• Recommender Systems
• Decision Theory
• Computational Sustainability

Background:

• Bundle recommendation seeks to group associated products, but modeling user utility is complex due to attribute interdependencies.
• Existing methods often fail to capture synergistic effects between product attributes, impacting recommendation quality.

Purpose of the Study:

• To introduce a novel preference-based bundle recommendation approach utilizing the Choquet integral.
• To formalize preferences for attribute coalitions, enabling recommendations that account for attribute synergies.
• To evaluate the effectiveness of the Choquet integral in improving environmental friendliness of recommended product bundles.

Main Methods:

• Developed a preference-based bundle recommendation framework employing the Choquet integral.
• Formalized preferences for coalitions of environmental attributes to capture interdependencies.
• Conducted experimental evaluations using a dataset of local food products from Northern Italy.
• Investigated preference elicitation strategies to acquire Choquet integral weights from user feedback.

Main Results:

• The Choquet integral effectively formalizes environmental friendliness in bundle recommendations.
• Standard weighted sum approaches recommended bundles with lower environmental friendliness, even when optimized for it.
• The proposed method demonstrates superior performance in recommending environmentally friendly product bundles.
• Preference elicitation strategies successfully acquired necessary weights, enabling optimal bundle recommendations with minimal user interaction.

Conclusions:

• The Choquet integral offers a robust method for bundle recommendation, particularly for capturing attribute synergies and environmental considerations.
• This approach enhances the accuracy and relevance of recommendations compared to traditional methods.
• Effective preference elicitation can significantly improve the personalization and efficiency of bundle recommendation systems.