AlignUSER: Human-Aligned LLM Agents via World Models for Recommender System Evaluation
Fuente:
arXiv
Guardado en:
| Autores principales: | , , , |
|---|---|
| Formato: | Preprint |
| Publicado: |
2026
|
| Materias: | |
| Acceso en línea: | |
| Etiquetas: |
Agregar Etiqueta
Sin Etiquetas, Sea el primero en etiquetar este registro!
|
| _version_ | 1866918269785473024 |
|---|---|
| author | Bougie, Nicolas Marconi, Gian Maria Yip, Tony Watanabe, Narimasa |
| author_facet | Bougie, Nicolas Marconi, Gian Maria Yip, Tony Watanabe, Narimasa |
| contents | 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. |
| format | Preprint |
| id |
arxiv_https___arxiv_org_abs_2601_00930 |
| institution | arXiv |
| publishDate | 2026 |
| record_format | arxiv |
| spellingShingle | AlignUSER: Human-Aligned LLM Agents via World Models for Recommender System Evaluation Bougie, Nicolas Marconi, Gian Maria Yip, Tony Watanabe, Narimasa Information Retrieval Artificial Intelligence 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. |
| title | AlignUSER: Human-Aligned LLM Agents via World Models for Recommender System Evaluation |
| topic | Information Retrieval Artificial Intelligence |
| url | https://arxiv.org/abs/2601.00930 |