Information Extraction for Integrating Multidisciplinary Articles to Co-construct Coastal Knowledge

Overview & Synopsis

Coastal areas, among the most dynamic and complex systems on Earth, face escalating environmental pressures due to anthropogenic activities and global change. The growth of coastal activities has generated competing interests and fragmentation in coastal development, leading to conflicts in land use and competition for natural resources. The rapid development of coastal regions has had substantial environmental consequences, including ecological disturbances, pollution, and increased vulnerability to natural hazards. The expansion of tourism and urbanization has further exacerbated these challenges, disproportionately affecting poor and vulnerable communities. Achieving a comprehensive understanding of the general functioning of the system requires a thorough integration of interdisciplinary knowledge. However, interdisciplinary connections are complex due to the incompatibility of inputs and outputs across different disciplines. This thesis lays the groundwork for a system designed to automatically construct an interdisciplinary Knowledge Base (KB) focused on the roles of various entities within littoral systems using Deep Learning. The key contributions include analyzing the evolution of coastal research from 1970 to 2023, investigating the application of the ARDI (Actor, Resource, Dynamic, Interaction) framework to guide the annotation strategy, employing Automatic Term Extraction (ATE) and end-to-end Document-level Relation Extraction (DocRE) to identify key terms and simultaneously extract entities and their relations from the dataset. The manuscript is organized into seven chapters, establishing the contextual framework and motivations underpinning the thesis, providing a comprehensive state-of-the-art review, presenting a scientometric analysis of coastal research output, delineating the dataset annotation process, describing the methodology and results of ATE, and introducing the model for the simultaneous extraction of entities and their relations. The thesis concludes with a discussion of the findings and their implications for future research and coastal management.