CUEPAQ: Visual Analytics and Linguistics for Capturing, Understanding, and Explaining Personalized Argument Quality

在我们的项目，视觉分析和语言学用于捕获，理解和解释个性化论元质量(CUEPAQ)中，我们结合视觉分析和计算语言学领域的方法来产生新的方法，根据不同语言分析层次上的不同度量来分析论元质量。基于这一分析，我们提供了所谓的偏好配置文件，使用户能够深入了解他们的个人辩论行为，并将其与其他用户的行为进行比较。本项目的主要目标是捕获、理解和解释辩论的感知质量。为此，我们收集了各种风格、内容和语义特征，这些特征影响了论点的框架和理解方式。这个项目的中心问题是这些元素是如何相互作用来产生被认为是高质量的论点的。在回答这个问题时，我们对论点质量的研究做出了贡献，这是一个用于论点评级和排名的可视化分析框架。该系统能够快速分析论点质量和论点的语言表达之间的交互作用。我们的框架提取偏好配置文件，这些配置文件通过指示特别影响用户参数评级的内容和文体特征来捕获用户的注释行为。这些偏好简档可能因用户而异，也可能因不同的用户组而异。为了解释这一点，我们在论元质量的分析中既包括了关于论元质量注释的专业知识，也包括了非专家用户评级的结果。根据论元之间的相对偏好比较，该系统提取了语言特征的模式，包括文体特征和解释性特征。这些功能预计会捕获用户的偏好，并因此反映在他们的评级行为中。这种外部化的知识是基于某些指导策略而可视化的，并允许用户和系统相互学习。由于系统可以跟踪不同用户的标注行为，这种自适应过程不仅允许用户了解自己的论证偏好，还可以将他们与系统中的其他用户进行比较，以及专家对高质量论证的意见。在这个项目中，我们使用计算语言方法来探索语言选择与用户和基于专家意见的系统对论点的排名之间的关系。具体地说，我们有助于统一标注与论元质量判断相关的论元的语言特征。

In our project, Visual Analytics and Linguistics for Capturing, Understanding, and Explaining Personalized Argument Quality (CUEPAQ) we combine methods from the fields of visual analytics and computational linguistic to generate new approaches for the analysis of argument quality in terms of various metrics on different levels of linguistic analysis. Based on this analysis, we provide, so-called, preference profiles, enabling users to gain insights into their personal argumentation behavior, as well as compare it to the behavior of other users.The main goal of this project is to capture, understand, and explain the perceived quality of arguments. To that end, we collect various stylistic, content, and semantic features that influence how arguments are framed and perceived. The central question of this project is how these elements interact to produce arguments that are perceived as high-quality.In answering this question, we contribute to the research on argument quality a visual analytics framework for the rating and ranking of arguments. The system enables rapid analysis of interactions between argument quality and the linguistic expression of an argument. Our framework extracts preference profiles, which capture the annotation behavior of users by indicating the content and stylistic features that particularly affect their rating of arguments. These preference profiles may vary from user to user, or across different user groups. To account for this, we include both expert knowledge on the annotation of argument quality, as well as results of non-expert user ratings in our analysis of argument quality. Based on relative preference comparisons between arguments, the system extracts patterns of linguistic features, both stylistic and interpretational. These features are expected to capture the users' preferences and would thus be reflected in their rating behavior. This externalized knowledge is visualized based on certain guidance strategies and allows both the user and the system to learn from each other. Since the system can keep track of the annotation behavior of different users, this co-adaptive process does not only allow a user to understand their own argumentation preferences but also to compare them with other users of the system, as well as the expert opinions on high-quality argumentation. In this project, we use computational linguistic methods to explore the relationship between linguistic choices and the ranking of arguments by users and systems based on expert-opinion. Concretely, we contribute to the uniform annotation of linguistic features of arguments that are relevant to the judgment of argument quality.