engleski

Computational Methods for Argumentation Mining of Claims in Internet Discussions

This thesis focuses on several tasks in argumentation mining of internet claims. Argumentation mining studies argumentation extraction, structuring, and analysis from text. Argumentation analysis is done in numerous applications, such as the analysis of scientific papers, policy decision making, or language learning through critical thinking development. With the increase of internet use, internet discussions are becoming a valuable source of argumentation. Claims constitute the building blocks of argumentation. This research proposes methods for mining topic-specific argumentative claim analysis in internet discussions. A framework and support for claim extraction and structuring and analysis is proposed. First, three pre-research studies based on unstructured approaches to argumentation mining are conducted. Then, based on the results of the pre-research studies, claim structuring is proposed. Claims are structured using a two-level ontology: an upper ontology and a domain ontology. The upper ontology is used to describe claim patterns. The domain ontology models domain-specific concepts. Structuring claims allows for higher quality claim analysis by relaxing the problem of language variance and allows for deriving implicit claims. Supervised machine learning methods are proposed to detect and structure claims from internet discussions. A method for claim analysis is proposed to analyse implicit claims of internet discussion participants.

argumentation mining ; natural language processing ; formal knowledge representation ; machine learning ; structured prediction

nije evidentirano