Aristotle University Of Thessaloniki (AUTH)
Machine learning techniques for fine-grained semantic indexing of biomedical literature
Abstract
dc:descriptionThis thesis focuses on machine-learning techniques for the automated indexing of biomedical literature with semantically related but distinct fine-grained concepts, such as disease subcategories, different proteins of the same family, or different species of organisms belonging to the same genus, which is beyond the current practice. Specifically, it investigates the challenges and potential in this unexplored and challenging task proposing new machine learning methods to tackle it. These methods complement and improve existing approaches for searching, retrieving, and integrating biomedical knowledge, identifying and addressing specific open issues, and making individual stages of biomedical research more efficient. In particular, this thesis formulates the task of Fine-Grained Semantic Indexing (FGSI) of the biomedical literature, hindered by the lack of ground truth labels. It introduces the Concept Occurrence (CO) heuristic as a strong and precise source of weak supervision and proposes two methods, Beyond MeSH and Deep Beyond MeSH (DBM), for training weakly supervised FGSI models that improve upon the weak supervision itself. In addition, it introduces the notion of the previous host (PH) and a novel framework for the conceptual provenance of new labels in the MeSH thesaurus. Based on this framework, it proposes the Retrospective Beyond MeSH method (RetroBM) for the development of large-scale ground-truth FGSI datasets. Finally, it also investigates the task of zero-shot relabeling of the weak labels, introducing the PN Relabeler method that learns from CO errors on past labels to improve the heuristic annotations on new unseen labels. The long-term goal of this thesis is to provide biomedical researchers with targeted and comprehensive access to the knowledge they need, accelerating their work and enabling them to make the most of the vast and ever-expanding biomedical literature. This way, we aim to pave the way to the discovery of new biomedical knowledge.
Degree
thesis:*- Grantor dc:publisher
- Aristotle University Of Thessaloniki (AUTH)
- Year dc:date
- 2024
Author and committee
dc:creator, dc:contributor.*- Authors dc:creator
-
- Nentidis, Anastasios
- Νεντίδης, Αναστάσιος
Subjects
dc:subject × 20- Σημασιολογική δεικτοδότηση
- Βιοϊατρική βιβλιογραφία
- Επικεφαλίδες ιατρικών θεμάτων (MeSH)
- Μηχανική μάθηση
- Ασθενώς επιβλεπόμενη μάθηση
- Βαθειά μάθηση
- Μάθηση χωρίς παραδείγματα
- Semantic indexing
- Biomedical literature
- Medical subject headings (MeSH)
- Machine learning
- Weakly supervised learning
- Deep learning
- Zero-shot learning
- Επιστήμη Ηλεκτρονικών Υπολογιστών και Πληροφορική
- Φυσικές Επιστήμες
- Τεχνητή νοημοσύνη
- Computer and Information Sciences
- Natural Sciences
- Artificial Intelligence
Rights
- Language dc:language
- eng
Identifiers
dc:identifier.*- Identifier
- 10.12681/eadd/57172