paperbot · PL 论文追踪

全部论文 待读 收藏

P2 Automated Testing of Conversational Agents with Chatbot Dōjō

Chatbot Dōjō:对话代理(chatbot)的自动化测试
ICSE Proceedings of the 2026 IEEE/ACM 48th International Conference on Software Engineering2026匹配度 2/10 · 论文关注聊天机器人黑盒端到端自动化测试、用户模拟器与测试DSL,属于软件测试与领域特定语言方向,与用户核心的静态分析、浮点/贝叶斯程序分析以及LLM与程序分析交叉方向基本无关;仅“对话Agent测试”与Agentic程序测试有微弱概念关联。引用 0 inbox
Iván Sotillo del Horno, Alejandro del Pozzo, Esther Guerra, Juan de Lara
crossref: 2026-09-12

AI 速览

一句话:提出黑盒端到端聊天机器人自动化测试 Web 平台 Chatbot Dōjō,含用户模拟器、测试 DSL 与探索器

对话代理(chatbot)以自然语言交互,可为订餐、客服等任务提供软件服务入口,其普及使可靠性成为关键需求。然而,现有工作缺少能适配多种 chatbot 技术的自动化测试技术。为此,本文提出 Chatbot Dōjō:一个将 chatbot 视为黑盒的端到端自动化测试 Web 平台。它由三部分组成:(i)用户模拟器,可通过测试配置定制生成对话、抽取对话中的相关数据,并检测崩溃或超时等问题;(ii)领域特定测试语言,用于定义分析对话的测试规则;(iii)chatbot 探索器,探测待测 chatbot 以生成配置,支持聚焦测试会话。平台已部署为公开服务并提供演示视频。

问题
对话代理日益普及且需可靠,但缺少能应对多种 chatbot 技术的自动化测试方法。
方法
提出 Chatbot Dōjō 平台,将 chatbot 作为黑盒进行端到端测试,结合可定制用户模拟器、领域特定测试语言与探索器。
评估
以公开部署的 Web 平台和演示视频展示功能;摘要未报告定量评估结果。
原文摘要(Abstract)

Conversational agents (or chatbots) interact in natural language, and may provide access to software services for specific tasks, like ordering food or obtaining customer support. Given their prominence and ubiquity, they need to be reliable. However, automated chatbot testing techniques, able to cope with the large variety of existing chatbot technologies, are missing. To fill this gap, we present Chatbot Dōjō: a web platform for end-to-end automated chatbot testing that treats chatbots as black boxes. It comprises: (i) a user simulator that can be customised via testing profiles to generate conversations with a chatbot, extract relevant data from conversations, and detect problems like crashes or timeouts; (ii) a domain-specific testing language to define testing rules to analyse conversations; and (iii) a chatbot explorer that probes a chatbot under test to generate profiles for focused testing sessions. The platform is deployed at http://miso.ii.uam.es:8081/. A demo video is available at https://youtu.be/dFsbN1FKGVg.

链接与引用

DOI 原文 · PDF(开放获取) ·

BibTeX
@inproceedings{paperbot11722,
  title = {Automated Testing of Conversational Agents with Chatbot Dōjō},
  author = {Iván Sotillo del Horno and Alejandro del Pozzo and Esther Guerra and Juan de Lara},
  booktitle = {ICSE},
  year = {2026},
  doi = {10.1145/3774748.3787604}
}