LLM-based E-commerce Navigating Agent for User Behavior Simulation

The Hong Kong University of Science and Technology
Department of Computer Science and Engineering

Final Year Thesis Oral Defense

Title: "LLM-based E-commerce Navigating Agent for User Behavior Simulation"

by

LAU Ching Ming Samuel

Abstract:

Simulating user behavior on E-commerce websites presents a significant 
challenge, as traditional approaches often rely on static simulation 
environments with rigid layouts and predefined product data. Although recent 
advancements in LLM agents have enabled more sophisticated interactions, 
these agents still struggle to operate effectively in dynamic, real-world 
platforms such as Amazon or eBay. To overcome these limitations, this thesis 
introduces an E-commerce web navigating agent framework that leverages vision 
language models to interpret user queries and interact with live websites. A 
benchmark is proposed to evaluate agents on diverse, realistic user intents, 
along with fine-grained evaluation metrics that assess both success and 
hallucination. Empirical evaluation across multiple architectures and LLM 
backbones shows our approach outperforms existing solutions, offering a more 
accurate and adaptable solution for real-world E-commerce navigation.


Date            : 2 May 2025 (Friday)

Time            : 13:50 - 14:30

Venue           : Room 5566 (near lifts 27/28), HKUST

Advisor         : Dr. SONG Yangqiu

2nd Reader      : Prof. DING Cunsheng