Andrew Qin and Landon Butler and Yigit Efe Erginbas and Kannan Ramchandran

EECS Department, University of California, Berkeley

Technical Report No. UCB/EECS-2025-118

May 16, 2025

http://www2.eecs.berkeley.edu/Pubs/TechRpts/2025/EECS-2025-118.pdf

In large-scale marketplaces, recommender systems are traditionally optimized from the platform’s perspective, aiming to maximize revenue or user purchase-rates. In this setting, due to the computational asymmetry, users are price-takers who must rely on recommender systems to suggest goods. We outline and study recommendation agents: systems designed to model and maximize an individual buyer’s utility. We apply candidate retrieval, sequential recommendation, and discrete choice modeling to capture contextual user desires and price sensitivities. Our system is evaluated on a semi-synthetic dataset based on e-commerce purchase behavior. We conclude with a discussion on the construction of markets over many recommendation agents, allowing for synchronous price negotiation, effectively simulating a logit-demand price competition setting, which opens opportunities for more efficient marketplace dynamics.

Advisors: Kannan Ramchandran

---

BibTeX citation:

@mastersthesis{Qin:EECS-2025-118,
    Author= {Qin, Andrew and Butler, Landon and Erginbas, Yigit Efe and Ramchandran, Kannan},
    Title= {Buyer-Side Recommendation Agents},
    School= {EECS Department, University of California, Berkeley},
    Year= {2025},
    Month= {May},
    Url= {http://www2.eecs.berkeley.edu/Pubs/TechRpts/2025/EECS-2025-118.html},
    Number= {UCB/EECS-2025-118},
    Abstract= {In large-scale marketplaces, recommender systems are traditionally optimized from the platform’s perspective, aiming to maximize revenue or user purchase-rates. In this setting, due to the computational asymmetry, users are price-takers who must rely on recommender systems to suggest goods. We outline and study recommendation agents: systems designed to model and maximize an individual buyer’s utility. We apply candidate retrieval, sequential recommendation, and discrete choice modeling to capture contextual user desires and price sensitivities. Our system is evaluated on a semi-synthetic dataset based on e-commerce purchase behavior. We conclude with a discussion on the construction of markets over many recommendation agents, allowing for synchronous price negotiation, effectively simulating a logit-demand price competition setting, which opens opportunities for more efficient marketplace dynamics.},
}

EndNote citation:

%0 Thesis
%A Qin, Andrew 
%A Butler, Landon 
%A Erginbas, Yigit Efe 
%A Ramchandran, Kannan 
%T Buyer-Side Recommendation Agents
%I EECS Department, University of California, Berkeley
%D 2025
%8 May 16
%@ UCB/EECS-2025-118
%U http://www2.eecs.berkeley.edu/Pubs/TechRpts/2025/EECS-2025-118.html
%F Qin:EECS-2025-118