UKP-SQuARE: A Software Platform for Question Answering Research

UKP-SQuARE：问答研究软件平台

• 批准号: 443179992
• 负责人: Professorin Dr. Iryna Gurevych
• 金额: --
• 依托单位: Fachgebiet Ubiquitäre Wissensverarbeitung
• 依托单位国家: 德国
• 项目类别: Research Grants
• 资助国家: 德国
• 项目状态: 未结题
• 来源: https://gepris.dfg.de/gepris/projekt/443179992?language=en
• 关键词: UKP; SQuARE; Software; Platform; Question

The number of use-cases that can be addressed by means of automatic question answering (QA) has exploded over the past years. End-to-end QA systems play a crucial role in enabling researchers to analyse the real-world performances of their models in these different scenarios. However, to date, QA systems and frameworks are limited to certain types of QA pipelines, and they are not reusable across diverse kinds of QA.In this project, we will address this issue by creating UKP-SQuARE, a reusable open-source QA platform that will allow researchers to seamlessly integrate novel approaches to different kinds of QA. UKP-SQuARE will include extensible components to retrieve web documents or knowledge-base information, a QA model repository to easily replace trained models during run-time, and a web-UI that can render explanations for model decisions. A QA skill system will finally enable researchers to integrate multiple interoperable approaches to different kinds of QA into one unified system. We will achieve this by consolidating the research prototypes from our DFG-funded project QA-EduInf for knowledge-base QA (KB-QA) and community-based QA (cQA). This will ensure the long-term preservation and the impact of our project results well beyond cQA and KB-QA.For the quality assurance of UKP-SQuARE, we will substantially involve the research community in the development process. We will moderate an open-source development community, provide detailed documentation and developer guides, and hold regular workshops with leading QA researchers. Furthermore, our public system demonstration will provide an extensive example of our platform in a real-world environment. Finally, we will integrate claim validation as a novel kind of QA in our platform to verify UKP-SQuARE’s extensibility, and its potential use in joint studies with other research fields such as social sciences.

可以通过自动问答（QA）解决的用例数量在过去几年中呈爆炸式增长。端到端QA系统在使研究人员能够分析其模型在这些不同场景中的真实世界性能方面发挥着至关重要的作用。然而，到目前为止，QA系统和框架仅限于某些类型的QA管道，并且它们不能在不同类型的QA中重用。在这个项目中，我们将通过创建UKP-SQuARE来解决这个问题，这是一个可重用的开源QA平台，可以让研究人员无缝集成不同类型的QA的新方法。UKP-SQuARE将包括可扩展的组件来检索Web文档或知识库信息，QA模型存储库可以在运行时轻松替换训练模型，以及可以为模型决策提供解释的Web UI。QA技能系统将最终使研究人员能够将多种可互操作的方法集成到一个统一的系统中。我们将通过巩固我们的DFG资助的项目QA-EduInf的知识库QA（KB-QA）和基于社区的QA（cQA）的研究原型来实现这一目标。 这将确保我们的项目结果的长期保存和影响远远超出cQA和KB-QA。为了保证UKP-SQuARE的质量，我们将在开发过程中充分参与研究界。我们将管理一个开源开发社区，提供详细的文档和开发人员指南，并与领先的QA研究人员定期举办研讨会。此外，我们的公共系统演示将提供我们的平台在真实世界环境中的广泛示例。最后，我们将在我们的平台中集成声明验证作为一种新型的QA，以验证UKP-SQuARE的可扩展性，以及它在与其他研究领域（如社会科学）联合研究中的潜在用途。