Advances in Deep Learning Methodologies have had a significant impact on the AI community. These innovations have made various tasks easier in industries like healthcare, social media, engineering, finance, and education. Large Language Models (LLMs) are one of the most notable deep learning inventions that have gained popularity for their incredible use cases. They imitate human behavior and utilize Natural Language Processing or Computer Vision to provide amazing solutions.
The application of Large Language Models in Ontology Engineering has been a topic of discussion. Ontology Engineering is a branch of knowledge engineering that deals with creating, building, curating, assessing, and maintaining ontologies. An ontology is a formal and precise specification of knowledge in a specific area, providing a systematic vocabulary of concepts, attributes, and relationships between them. This enables shared understanding between humans and machines.
However, a challenge arises when it comes to integrating well-known ontology APIs like the OWL API and Jena, which are mostly Java-based, with deep learning frameworks like PyTorch and Tensorflow, which are primarily developed for Python programming. To address this challenge, a team of researchers has introduced DeepOnto. It is a Python package specifically designed for ontology engineering that allows seamless integration of these frameworks and APIs.
The DeepOnto package offers comprehensive, general, and Python-friendly support for deep learning-based ontology engineering. It includes an ontology processing module that supports basic operations like loading, saving, querying entities, and modifying entities and axioms. It also includes advanced functions like reasoning and verbalization. DeepOnto provides tools and resources for ontology alignment, completion, and ontology-based language model probing.
The team has chosen the OWL API as the backend dependency for DeepOnto due to its stability, reliability, and widespread adoption in notable projects and tools. PyTorch serves as the foundation for DeepOnto’s deep learning dependencies because of its dynamic computing graph, which allows for runtime adjustment of the model’s architecture, providing flexibility and usability.
The practical utility of DeepOnto has been demonstrated through two use cases. In one use case, it has been used for ontology engineering in the framework of Digital Health Coaching at Samsung Research UK. In another use case, DeepOnto has been used for aligning and completing biomedical ontologies using deep learning techniques in the Ontology Alignment Evaluation Initiative (OAEI)’s Bio-ML track.
In conclusion, DeepOnto is a powerful package for ontology engineering and a valuable addition to developments in the field of Artificial Intelligence. It provides a flexible and expandable interface for future implementations and projects. For more information, you can check out the paper and GitHub link. You can also join the ML subreddit, Discord channel, and email newsletter to stay updated on the latest AI research news and projects.