PsyBer-Agent Released: Revolutionizing User Behavior Modeling in Recommender Systems with Psychology-Driven Simulation
Summary:Recommender systems require simulators that maintain realism over extended interaction cycles, but current LLM-based agents often fail to capture underlying user dynamics such as fatigue, affect, and shifting preferences. PsyBer-Agent addresses this with its Psy-Engine, which models user behavior through three evolving latent states and aligns simulated trajectories with real interaction logs via unbalanced Gromov-Wasserstein optimal transport. This approach provides structured supervision for l
Technical Mechanism Analysis
The core of PsyBer-Agent is the Psy-Engine, which models user behavior through three evolving latent states:
- Fatigue State: Captures the user's fatigue level over time.
- Affect State: Models the user's emotional fluctuations.
- Preference State: Reflects the dynamic shifts in user preferences.
These states are aligned with reference trajectories from real interaction logs using the unbalanced Gromov-Wasserstein optimal transport algorithm, providing precise simulation behavior. This approach overcomes the limitations of traditional LLM-based agents in capturing long-term user dynamics while avoiding the computational burden of fine-tuning the LLM.
Engineering Trade-offs and Performance
PsyBer-Agent optimizes in several aspects:
- Computational Efficiency: By providing lightweight policy guidance, it avoids frequent LLM fine-tuning, reducing computational costs.
- Simulation Fidelity: In tests on the WebSim platform and PsyBer Benchmark, PsyBer-Agent shows significantly lower behavioral discrepancy compared to prompt-only simulators.
- Concurrent Performance: It maintains stable performance under large-scale concurrent simulation for up to 5 million queued tasks under bounded concurrency.
Developer Implementation and Deployment Recommendations
PsyBer-Agent is suitable for recommender system simulation scenarios that require long-term realism, especially in areas involving dynamic user behavior changes. Here are some deployment recommendations:
- Data Preparation: Ensure sufficient real interaction logs are available for training and calibrating the simulator.
- Model Tuning: Adjust the number and definition of latent states according to the specific application scenario to better capture user behavior.
- Benchmarking: Use the WebSim platform and PsyBer Benchmark for model evaluation to ensure the fidelity and robustness of simulated behaviors.
Conclusion
PsyBer-Agent, with its innovative psychology-driven simulation mechanism, fills the gap in user dynamic behavior modeling for existing LLM-based agents. Its advantages in computational efficiency, simulation fidelity, and concurrent performance make it a powerful tool in the field of recommender system simulation.
— END —Source: Hugging Face Daily Papers (2026-09-26)
Tags: #PsyBer #Recommender Systems #Psychology-Driven #Simulator #User Behavior Modeling
Editorial & Fact-Checking Note: This article is compiled from primary research, official release documentation, and source papers by the ZICQ Newsroom pipeline with automated entity verification and human editorial review. If you notice any technical inaccuracy, please submit a correction via our corrections policy or email our editorial desk directly.
Community Comments