AlignUSER: Human-Aligned LLM Agents via World Models for Recommender System Evaluation

TL;DR

AlignUSER presents a world-model-guided framework for learning human-aligned LLM agents to evaluate recommender systems. By pretraining on next-state prediction and applying counterfactual reasoning against human demonstrations, the method yields agents whose actions and reasoning align with human personas, improving micro- and macro-level evaluation signals. The paper demonstrates that AlignUSER and AlignUSER+ achieve closer human alignment, better rating predictions, and stronger proxy correlations with online A/B tests across several datasets. These results suggest synthetic user agents can provide scalable, privacy-preserving, and interpretable evaluation of RS performance while offering insights into user behavior and evaluation dynamics.

Abstract

Evaluating recommender systems remains challenging due to the gap between offline metrics and real user behavior, as well as the scarcity of interaction data. Recent work explores large language model (LLM) agents as synthetic users, yet they typically rely on few-shot prompting, which yields a shallow understanding of the environment and limits their ability to faithfully reproduce user actions. We introduce AlignUSER, a framework that learns world-model-driven agents from human interactions. Given rollout sequences of actions and states, we formalize world modeling as a next state prediction task that helps the agent internalize the environment. To align actions with human personas, we generate counterfactual trajectories around demonstrations and prompt the LLM to compare its decisions with human choices, identify suboptimal actions, and extract lessons. The learned policy is then used to drive agent interactions with the recommender system. We evaluate AlignUSER across multiple datasets and demonstrate closer alignment with genuine humans than prior work, both at the micro and macro levels.

Figures (6)

• Figure 1: The AlignUSER framework for evaluating a recommender system by implicitly modeling a world model and exploring alternative scenarios.
• Figure 2: Counterfactual reflection from counterfactual trajectories.
• Figure 3: Spearman correlation between estimated and actual engagement metrics. Higher values indicate better alignment with ground-truth metrics.
• Figure 4: Comparison of RMSE values for original (dark colors) and hallucination-affected (light colors) models for the rating task on MovieLens.
• Figure 5: Comparison of rating distributions between ground-truth and human proxies.