Argumentation Analysis for the Web

点击翻译按钮获取中文摘要

Argumentation mining deals with the automatic identification of arguments and their relations from natural language text. This research project targets at the specific challenges of argumentation mining for the web. We seek to establish foundations of algorithms that (1) robustly apply to various forms of web argumentation, (2) efficiently leverage the scale of the web, and (3) complement argumentation mining with an argumentation analysis to effectively assess important quality dimensions.The rationale of the planned project is that people compare arguments in many decision-making situations, e.g., when buying products or when forming opinions on political controversies. Nowadays, the richest and most up-to-date argument source is the web. However, searching for arguments on the web is challenging, as dozens of web pages need to be read through in order to identify and relate the relevant arguments. State-of-the-art research on argumentation mining tackles the identification and relation of arguments within a particular domain, but it does not suffice to successfully mine argumentation on the web. The web contains numerous texts with monological argumentation (like opinionated news articles) and dialogical argumentation (like the discussions below articles) from various domains. Existing argumentation mining approaches build upon specific models from argumentation theory that do not cover this variety. The approaches rely on manually annotated samples of text, which cannot be obtained for all domains due to the scale of the web. Moreover, they disregard that, especially on the web, the quality of argumentation strongly varies with respect to several dimensions, such as clarity, coherence, or the presence of fallacies. In this project, we aim to evolve models from argumentation theory to make them comply with major forms of web argumentation. Then, we will create annotated corpora with tens of thousands of argumentative web texts from different domains. To keep the annotation effort tractable, we plan to employ distant supervision and games with a purpose. Based on the corpora, we will develop and evaluate novel algorithms that mine web argumentation and that learn patterns in it, which affect measurable quality dimensions. Domain adaptation techniques, among others, will help to cope with the variety of the web. While the size of the corpora raises the need for efficiency, it will also bring unprecedented statistical insights into web argumentation across domains.We expect to obtain new knowledge about common, good, and bad ways in which people argue on the web, thereby bridging the existing gap between theory and the practical use of argumentation. The created corpora will serve as valuable resources for other researchers, and the algorithms will be able to mine argumentation that meets specific quality constraints from a variety of web texts. We believe that leveraging such argumentation will shape the future of the web search.