In Search of a Common Thread: Enhancing the LBD Workflow with a view to its Widespread Applicability

Abstract

Literature-Based Discovery (LBD) research focuses on discovering implicit knowledge linkages in existing scientific literature to provide impetus to innovation and research productivity. Despite significant advancements in LBD research, previous studies contain several open problems and shortcomings that are hindering its progress. The overarching goal of this thesis is to address these issues, not only to enhance the discovery component of LBD, but also to shed light on new directions that can further strengthen the existing understanding of the LBD work ow. In accordance with this goal, the thesis aims to enhance the LBD work ow with a view to ensuring its widespread applicability. The goal of widespread applicability is twofold. Firstly, it relates to the adaptability of the proposed solutions to a diverse range of problem settings. These problem settings are not necessarily application areas that are closely related to the LBD context, but could include a wide range of problems beyond the typical scope of LBD, which has traditionally been applied to scientific literature. Adapting the LBD work ow to problems outside the typical scope of LBD is a worthwhile goal, since the intrinsic objective of LBD research, which is discovering novel linkages in text corpora is valid across a vast range of problem settings. Secondly, the idea of widespread applicability also denotes the capability of the proposed solutions to be executed in new environments. These `new environments' are various academic disciplines (i.e., cross-domain knowledge discovery) and publication languages (i.e., cross-lingual knowledge discovery). The application of LBD models to new environments is timely, since the massive growth of the scientific literature has engendered huge challenges to academics, irrespective of their domain. This thesis is divided into five main research objectives that address the following topics: literature synthesis, the input component, the discovery component, reusability, and portability. The objective of the literature synthesis is to address the gaps in existing LBD reviews by conducting the rst systematic literature review. The input component section aims to provide generalised insights on the suitability of various input types in the LBD work ow, focusing on their role and potential impact on the information retrieval cycle of LBD. The discovery component section aims to intermingle two research directions that have been under-investigated in the LBD literature, `modern word embedding techniques' and `temporal dimension' by proposing diachronic semantic inferences. Their potential positive in uence in knowledge discovery is veri ed through both direct and indirect uses. The reusability section aims to present a new, distinct viewpoint on these LBD models by verifying their reusability in a timely application area using a methodical reuse plan. The last section, portability, proposes an interdisciplinary LBD framework that can be applied to new environments. While highly cost-e cient and easily pluggable, this framework also gives rise to a new perspective on knowledge discovery through its generalisable capabilities. Succinctly, this thesis presents novel and distinct viewpoints to accomplish five main research objectives, enhancing the existing understanding of the LBD work ow. The thesis offers new insights which future LBD research could further explore and expand to create more eficient, widely applicable LBD models to enable broader community benefits.