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

arXiv 26.1 2026 60 method, application

TLDR

AlignUSER uses world models and counterfactual trajectories to align LLM agents with human behavior for recommender system evaluation.

Reasoning

The paper introduces a novel framework that combines world modeling with human alignment, addressing limitations of few-shot prompting. Strengths include explicit next-state prediction and counterfactual learning, while weaknesses are not evident from the abstract alone. The evaluation on multiple datasets supports real-world applicability.

Read-first score

Read-first score 60, weighted from topical fit, citation, graph, method, reproducibility, and recency signals. Original total remains 38.

Recency 8%

100

Uses a gentle age decay so recent papers surface without erasing older foundations. 2026

Methodology quality 25%

Screens visible abstract and analysis fields for experiment, dataset, baseline, metric, and limitation evidence. markers=dataset,evaluation,metric

Topical relevance 42%

54.3

Uses existing LLM keyword relevance scores normalized to 0-100. world model,world simulator,generative world model,interactive world model,video world model,world dynamics prediction,model-based reinforcement learning world model

Reproducibility 25%

Screens links and visible text for paper, code, dataset, artifact, and repository signals. pdf=True; code=False; dataset=False; markers=dataset,reproduce

Field roles

FrontierMethodology anchor

Rank sensitivity

Stability: volatile; rank range: 344.

Keyword Scores

world model

world dynamics prediction

interactive world model

model-based reinforcement learning world model

world simulator

generative world model

video world model

Deep Analysis

Innovations

World-model-driven agent learning from human interactions for recommender system evaluation
Formalizing world modeling as a next state prediction task from rollout sequences
Using counterfactual trajectories around demonstrations to align agent actions with human personas via LLM comparison and lesson extraction

Methodology

AlignUSER formalizes world modeling as a next state prediction task from rollout sequences of actions and states, enabling the agent to internalize the environment. To align actions with human personas, it generates counterfactual trajectories around demonstrations and prompts the LLM to compare its decisions with human choices, identify suboptimal actions, and extract lessons. The learned policy then drives agent interactions with the recommender system.

Key Results

AlignUSER is evaluated across multiple datasets and demonstrates closer alignment with genuine humans than prior work, both at the micro and macro levels.

Resources

Paper