NEU-HAI/OPeRA

OPeRA数据集是从真实的人类用户在线购物过程中收集的观察、个性、理据和动作记录的数据集。它提供了丰富的、时间对齐的用户网页浏览行为日志，并补充了自我报告的理据和详细的自我报告个性配置文件。与之前的数据库不同，OPeRA不仅捕捉了用户的行为，还捕捉了他们为什么这么做，从而能够更深入地了解决策过程。为了确保用户隐私和数据安全，发布的数据库经过了严格的过滤和去标识处理。该数据集包含来自51个独特用户的692个购物会话，28,904个<观察，动作>对，以及604个人类注释的理据。OPeRA作为评估LLM代理在个性化和可验证的用户行为模拟上的第一个基准数据库。

OPeRA is a dataset of Observation, Persona, Rationale, and Action collected from real human users during online shopping. It provides rich, time-aligned logs of users web browsing behavior, completed with self-reported rationales and detailed self-reported persona profiles. Unlike prior datasets, OPeRA captures not only what users do but also why they do it, enabling deeper insights into decision-making processes. To ensure user privacy and data safety, the released dataset has undergone rigorous filtering and de-identification. The dataset contains 692 shopping sessions from 51 unique users, 28,904 `<observation, action>` pairs, and 604 human-annotated rationales. OPeRA serves as the first benchmark dataset for evaluating LLM agents on personalized and verifiable user behavior simulation.