Deep Learning and Ontology Engineering Python Package

Advances in Deep Learning Methodologies are greatly impacting the Artificial Intelligence community. With some great innovations and developments, a number of tasks are getting easier. Deep Learning techniques are being widely used in almost every industry, be it healthcare, social media, engineering, finance, or education. One of the best deep learning inventions is Large Language Models (LLMs), which recently got popular and are mostly in the headlines for their incredible use cases. These models imitate humans and, by utilizing the power of Natural Language Processing or Computer Vision, demonstrate some amazing solutions. 

The application of Large Language Models in the field of Ontology Engineering has been the topic of discussion ever since. Ontology engineering is a branch of knowledge engineering that is concerned with the creation, building, curation, assessment, and upkeep of ontologies. An ontology is basically a formal and precise specification of knowledge within a particular area that offers a systematic vocabulary of concepts and attributes, along with the relationships between them, in order to enable a shared understanding of semantics between humans and machines. 

Well-known ontology APIs like the OWL API and Jena are mostly Java-based, while deep learning frameworks like PyTorch and Tensorflow are developed generally for Python programming. This comes as a challenge to address which, a team of researchers has introduced DeepOnto, a Python package developed specifically for ontology engineering that enables seamless integration of the frameworks and APIs. 

DeepOnto package provides comprehensive, general, and Python-friendly support for deep learning-based ontology engineering, and it consists of an ontology processing module as the foundation which supports basic operations such as loading, saving, querying entities, modifying entities and axioms, and advanced functions like reasoning and verbalization. It also includes 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. This is because of the characteristics of the API, such as its stability, reliability, and widespread adoption in notable projects and tools such as ROBOT and HermiT. PyTorch is the foundation for DeepOnto’s deep learning dependencies due to its dynamic computing graph, which permits runtime adjustment of the model’s architecture, offering flexibility and usability. Huggingface’s Transformers library has been used for language model applications, and the OpenPrompt library has been used to support the prompt learning paradigm, which is a crucial underpinning for big language models like ChatGPT.

DeepOnto’s basic ontology processing module is made up of a number of parts, each of which performs a particular task ? First is Ontology, DeepOnto’s base class that offers the fundamental methods for viewing and changing an ontology. Second, is ontology reasoning, which is used for conducting reasoning activities, followed by Ontology pruning in which an ontology is taken and a scalable subset is extracted depending on particular criteria, such as semantic kinds. Lastly, Ontology Verbalization is there which improves the ontology’s accessibility and aids in a variety of ontology engineering activities by verbalizing ontology elements into natural language text.

The team has demonstrated the practical utility of DeepOnto with the help of two use cases. In the first use-case, DeepOnto has been used to help ontology engineering tasks within the framework of Digital Health Coaching at Samsung Research UK. The Ontology Alignment Evaluation Initiative (OAEI)’s Bio-ML track is the second use-case, where DeepOnto has been used to align and finish biomedical ontologies using deep learning techniques.

In conclusion, DeepOnto is a strong package for ontology engineering and is a great addition to the developments in the field of Artificial Intelligence. For future implementations and projects like logic embeddings and the discovery and introduction of new concepts, DeepOnto provides a flexible and expandable interface.