A Learning Framework for Personalized Random Utility Maximization (RUM) Modeling of User Behavior

Area of Science:

• Behavioral economics
• Machine learning
• Data science

Background:

• Personalized services in health and transportation apps require understanding user behavior.
• Random utility maximization (RUM) models struggle with fragmented individual data.
• Existing models face challenges in capturing heterogeneous user preferences with limited data.

Purpose of the Study:

• To develop a framework for modeling individual user preferences from limited and fragmented data.
• To address the limitations of RUM models in personalized service applications.
• To propose an extension for handling uneven canonical structures in real-world scenarios.

Main Methods:

• Utilized concepts of canonical structure and membership vectors from collaborative learning.
• Developed an extended collaborative learning framework with pairwise-fusion regularization.
• Designed computationally efficient algorithms for optimization challenges.
• Applied methods to smart transportation demand management (TDM).

Main Results:

• The proposed framework effectively models heterogeneous populations with insufficient individual data.
• Pairwise-fusion regularization aids knowledge discovery in applications with uneven canonical structures.
• Demonstrated effectiveness through extensive simulations and a real-world TDM application.

Conclusions:

• The novel collaborative learning framework significantly improves the estimation of individual preferences.
• The methods are robust and effective for personalized service applications with data limitations.
• The approach offers a powerful tool for understanding and modeling complex user behavior.