RecVerse：忠实模拟人类购物行为的 GUI 智能体

AI 摘要

arXiv 论文提出 RecVerse，一种基于 GUI 的购物行为模拟智能体，通过截图感知页面并生成多轮轨迹。它采用认知启发的分层记忆（工作记忆、情景记忆、偏好记忆）解决长会话记忆挑战，并用轨迹级强化学习优化整体行为分布和购物意图。实验表明 RecVerse 在行为保真度和意图一致性上优于现有基线，并发布了 USB 基准数据集。

正文节选

Towards Faithful Simulation of Human Shopping BehaviorCCS: Information systems Recommender systemsCCS: Computing methodologies Reinforcement learningCCS: Computing methodologies Computer vision Abstract. Simulating realistic user shopping behavior underpins offline evaluation and reinforcement learning in e-commerce scenarios. While recent LLM- and VLM-based simulators have made encouraging progress, reproducing a real browsing session remains difficult for two reasons. (i) Memory Challenge: a s

发布时间：2026-08-24 12:00

抓取时间：2026-08-24 12:32

来源机构：arXiv