IDENTIFYING ARGUMENTATIVE CLAIMS IN BIOMEDICAL RESEARCH ARTICLES
Abstract
The proliferation of biomedical research literature has led to an urgent need for efficient methods to navigate and extract relevant claims from the extensive corpus of research articles. Manually identifying argumentative claims within these articles can be labor-intensive and time-consuming. This research addresses the complexities associated with claim detection in biomedical literature by proposing a novel approach that harnesses pre-trained models and employs titles as potential indicators for identifying argumentative claims across various sections of research articles. A notable contribution of this research is the formulation of a balanced subset from the MeSHup dataset, specifically targeting articles that contain verbs in their titles. Additionally, this study introduces an alternative methodology that involves fine-tuning biomedical language models to enhance the precision of claim detection in the text of research articles. The efficacy of these methodologies was rigorously assessed through the use of manual annotations conducted by trained annotators. As part of this research, we also develop a new dataset for biomedical argumentative claim detection using MeSHup articles. Ultimately, the results of this thesis will help to streamline literature reviews, facilitating the extraction of critical insights from scientific texts with heightened accuracy and efficiency, thereby advancing the understanding and application of biomedical research.