Team-ACE/ProActEval
收藏资源简介:
ProActEval是一个用于评估主动式AI代理的数据集,包含200个英文场景级示例。每个示例描述了一个合成用户场景,其中包含结构化事实、潜在或显性用户需求、揭示组和模拟器配置,旨在评估助手是否能预见和准备未来用户需求。数据集使用虚构的人物、组织、地点和任务上下文,不包含真实用户记录或个人数据。数据以JSONL格式提供,包括场景ID、领域、描述、用户档案、事实表、用户需求等字段。该数据集为合成基准,专为研究主动代理、代理记忆、空闲时间推理和个性化协助而设计,主要用于评估而非模型训练。
ProActEval is an evaluation dataset for proactive AI agents. Each example describes a synthetic user scenario with structured facts, latent or explicit user needs, reveal groups, and simulator configuration for evaluating whether an assistant can anticipate and prepare for future user needs. The dataset contains 200 scenario-level examples in English. All scenarios use fictional people, organizations, locations, and task contexts; the dataset does not contain real user records or real personal data. It is provided in JSONL format with fields such as scenario_id, domain, description, user_profile, fact_sheet, user_needs, reveal_groups, simulator_config, and metadata. The dataset is synthetic and designed for research on proactive agents, agent memory, idle-time reasoning, personalized assistance, and benchmarked anticipation of user needs, primarily intended for evaluation rather than model training.




