📄 中文摘要
在物联网、可观察性、电信和网络安全等多个领域，越来越多地采用对话式数据分析代理，使用户能够与数据进行交互以提取洞察。这些数据分析代理基于时间序列数据模型，例如来自传感器的测量值或监控用户点击和产品分析中的行为事件。对六种流行的数据分析代理（包括开源和专有）进行评估时，发现它们在状态保持和特定事件查询方面存在不足。现有评估中观察到两个主要的表达能力缺口：领域定制的数据集和领域特定的查询类型。因此，研究提出了一种方法，帮助从业者生成定制化和富有表现力的评估，以满足特定领域的需求。
使用 AgentFuel 生成可表达和可定制的时间序列数据分析代理评估
📄 English Summary
Generating Expressive and Customizable Evals for Timeseries Data Analysis Agents with AgentFuel
The research highlights the increasing adoption of conversational data analysis agents across various domains such as IoT, observability, telecommunications, and cybersecurity, allowing users to interact with their data for insights. These agents operate on timeseries data models, including sensor measurements and event monitoring in product analytics. An evaluation of six popular data analysis agents, both open-source and proprietary, reveals their shortcomings in handling stateful and incident-specific queries. Two significant expressivity gaps are identified in existing evaluations: the lack of domain-customized datasets and domain-specific query types. This study proposes a method to enable practitioners to generate customized and expressive evaluations tailored to the specific needs of their domains.