Intelligent knowledge integration from scientific publications on urban ecosystem services

Abstract

Integration of knowledge from scientific literature is increasingly hindered by knowledge fragmentation, due to the heavy reliance on manual integration and the incompatibility between integration results. When the scientific topics to integrate have a complex disciplinary background, the frag mentation problem can be even worse, such as urbanization and ecosystem services. To mitigate fragmentation, it is necessary to reduce the reliance on manual efforts while improving the compatibility of knowledge representation, which are the focuses of scientific relation extraction and knowledge graphs, respectively. However, integrating scientific relation extraction and knowledge graphs remains a challenge due to the lack of annotations and the inflexibility of constraining the results of the extraction based on knowledge graphs.

This thesis aims at mitigating fragmentation by increasing the flexibility of the integration of knowledge from the scientific literature on urbanization and ecosystem services, which is investigated in two main research questions: how to design a knowledge graph schema that supports an integration-friendly construction of knowledge graphs and how to evaluate and perform the integration of knowledge graphs. Methodologically, the thesis explores the integration between implicit knowledge graphsen coded in the models for scientific relation extraction and explicit knowledge graphs derived from the literature and external knowledge bases. A suite of methods for determining the integration focus is proposed, based on coarsening and prompting the paths in dependency knowledge graphs. Without data that directly pair the integration candidates, an integration frame work based on clustering completeness is proposed that supports one-to multiple relation alignment. Building on this framework, a neural research registration approach is proposed that combines bibliometric analysis with the completeness-based alignment to register studies on urbanization and ecosystem services.

Overall, this thesis advances scientific knowledge integration by highlighting the shift from entity-focused to relation-focused integration and relieving the constraints on the integration in terms of input context and integration data.

Date of Award	18 Jul 2026
Original language	English
Awarding Institution	University of Nottingham
Supervisor	Xiangjian He (Supervisor), Yaoyang Xu (Supervisor) & Simon Gosling (Supervisor)