PsyBer-Agent: A Psychology-Driven User Agent for Recommender Systems
Evaluating recommender systems before deployment requires simulators that remain realistic over long, closed-loop interaction cycles. Current LLM-based agents generally function as surface imitators that generate realistic short-term responses but fail to capture the underlying user dynamics such…
The article introduces PsyBer-Agent, a psychology-driven simulator designed to evaluate recommender systems by modeling user dynamics such as fatigue and shifting preferences through a novel framework.
What happened
The authors developed PsyBer-Agent to address limitations in existing LLM-based simulators for recommender systems. The approach models user psychology through latent states and uses optimal transport for calibration, enabling more accurate long-term simulations.
Why it matters
This work improves the evaluation of recommender systems by capturing realistic user behavior over extended interactions, which is critical for deploying reliable AI-driven recommendation technologies.
- Psychology-driven models improve the realism of user simulations in recommender systems.
- Optimal transport techniques provide structured supervision without fine-tuning large language models.
- The PsyBer Benchmark and WebSim platform offer standardized tools for evaluating AI-driven recommendation systems.
The bigger picture
The framework enables more accurate and scalable evaluation of recommender systems, reducing discrepancies between simulated and real-world user behavior while maintaining efficiency in large-scale simulations.
PsyBer-Agent advances the evaluation of recommender systems by integrating psychological principles and efficient simulation techniques, offering a scalable solution for testing AI-driven user interactions.
Technical details
- Psy-Engine: Three model-defined latent states representing user dynamics (e.g., fatigue, affect).
- Unbalanced Gromov-Wasserstein optimal transport: Aligns real interaction logs with simulated latent-state trajectories.
- WebSim: Multimodal simulation platform for testing recommender systems.
- PsyBer Benchmark: Includes seven recommender backbones and scenarios for evaluating behavioral consistency.
FAQ
What is PsyBer-Agent's primary innovation?
It introduces a psychology-inspired simulator that models user dynamics through latent states and uses optimal transport for calibration, addressing limitations of existing LLM-based agents.
How does PsyBer-Agent improve upon current simulators?
By capturing long-term user behavior (e.g., fatigue, shifting preferences) and reducing behavioral discrepancies through structured supervision without requiring LLM fine-tuning.
What tools are included in the PsyBer framework?
WebSim for multimodal simulations and the PsyBer Benchmark, which contains seven recommender backbones and scenarios for evaluating behavioral fidelity and robustness.
This summary and analysis were generated with AI from the original article and may contain errors. Read the original for the full story.
Read the full story
Originally published by arXiv - cs.CL (NLP) on Oct 14, 2026.
Comments
Sign in to join the discussion.