EvolutionAgent: A Large Model-Based Framework for User Behavior Modeling and Self-Evolving Intelligent Agents

Abstract

Large Language Models (LLMs) have demonstrated remarkable capabilities in simulating user behavior, offering significant potential for user modeling, behavior analysis, interest matching, and the extraction of unstructured features. However, the process of user simulation necessitates the identification of both unstructured and structured features, as well as the formulation of multi-stages workflow, which remains a challenging and labor-intensive task. To address this, we propose EvolutionAgent , a novel framework grounded in a discretized material repository, which employs self-reflective and evolutionary mechanisms to automate the search for unstructured features and the generation of simulation workflow. This framework establishes an efficient and robust simulation mechanism, achieving state-of-the-art performance across three distinct datasets. Notably, EvolutionAgent exhibits exceptional robustness, even in scenarios with limited historical data for users and products, underscoring its adaptability and reliability.