Dissertação

Word-embedding-based string similarity in ontology matching EVALUATED

Ontology matching, a pivotal process in knowledge integration, seeks to align concepts from diverse sources. However, traditional methods often struggle with capturing nuanced semantic relationships, relying heavily on syntactical similarities. This is where the potential of Large Language Models (LLMs) comes into play. LLMs, like BERT, offer contextual understanding, enabling them to grasp intricate semantic nuances beyond string similarity. Moreover, the concept of knowledge injection enriches the model’s comprehension. By incorporating domain specific knowledge like hierarchical parent names or synonyms, the model develops an enhanced comprehension of relationships within ontologies. This thesis uniquely combines the power of fine-tuned LLMs and knowledge injection to enhance ontology matching accuracy. The results showcase advancements in ontology alignment, outperforming conventional techniques. This study pioneers a refined method that not only contributes to ontology matching but also sets the stage for more accurate and contextually aware information integration. By leveraging the potential of LLMs and enhancing domain understanding, this approach offers a promising way forward in knowledge integration for key domains such as biomedicine.
ontology matching, knowledge integration, Large Language Models (LLMs), biomedical ontology, information integration, nuanced semantics.

setembro 13, 2023, 11:0

Publicação

Obra sujeita a Direitos de Autor

Orientação

ORIENTADOR

Daniel Pedro de Jesus Faria

Departamento de Engenharia Informática (DEI)

Professor Auxiliar