Generating Linguistic Insights in Question Classification throughCombining Explainable Machine Learning and Visualization

通过结合可解释的机器学习和可视化来生成问题分类中的语言见解

• 批准号: 276395906
• 负责人: Professorin Dr. Miriam Butt
• 金额: --
• 依托单位: Arbeitsgruppe Visual Computing
• 依托单位国家: 德国
• 项目类别: Research Units
• 财政年份: 2016
• 资助国家: 德国
• 起止时间: 2015-12-31 至 2022-12-31
• 项目状态: 已结题
• 来源: https://gepris.dfg.de/gepris/projekt/276395906?language=en
• 关键词: Generating; Linguistic; Insights; Question; Classification

This project tackles the question classification task: the task of automatically distinguishing between different types of canonical and non-canonical questions. The project encompasses three parts. One part focuses on the extraction and collection of linguistic information to be used as features for Machine Learning models (ML) and Visual Analytics (VA) techniques for the classification task. The second part includes VA techniques for the interactive adjustment of the ML models to improve question classification. Through an interactive manipulation of the model's visual representation we enable the user to interactively adapt and improve the learned model. The third part deals with making the ML techniques transparent in order to generate additional linguistic insights for question classification. We aim to develop novel methods to communicate decisions made by a ML model and provide linguistic insights for the task at hand. By pursuing these goals we are contributing to the research of questions by developing novel tools and methodology within computational linguistics and VA for the analysis and classification of question types. At the same time, we interact with other projects of the Research Unit in terms of carrying the possibilities opened up by LingVis (Visualization for Linguistics) into individual projects and applying them towards the work pursued by these projects. This project benefits from a feedback cycle in that it can integrate insights on the structure of questions coming from other projects into its own classificatory work and can in turn produce results that can then be carried into the other projects.

这个项目处理问题分类任务：自动区分不同类型的规范和非规范问题的任务。该项目包括三个部分。一部分侧重于语言信息的提取和收集，这些信息将用作机器学习模型（ML）和视觉分析（VA）技术用于分类任务的特征。第二部分包括用于交互式调整ML模型的VA技术，以改进问题分类。通过对模型可视化表示的交互式操作，我们使用户能够交互式地适应和改进学习到的模型。第三部分涉及使机器学习技术透明，以便为问题分类生成额外的语言见解。我们的目标是开发新的方法来传达由ML模型做出的决策，并为手头的任务提供语言见解。通过追求这些目标，我们正在通过在计算语言学和VA中开发用于分析和分类问题类型的新工具和方法，为问题的研究做出贡献。与此同时，我们与研究部门的其他项目进行互动，将LingVis（语言学可视化）所带来的可能性应用到各个项目中，并将其应用到这些项目所追求的工作中。该项目受益于反馈循环，因为它可以将来自其他项目的问题结构的见解集成到自己的分类工作中，并且可以反过来产生可用于其他项目的结果。