Ontology-based Inference and Learning

Introduction:

In AI, ontology is the formal specification of the vocabulary and relationships of the concepts. A lot of ontologies have been developed to describe the semantics of data resources in various domains, for example, the ontologies for knowledge bases and the emerging Semantic Web. The research related on ontology includes ontology representation, ontology inference, ontology mapping/merging and ontology translation etc.

Ontologies can be represented by some formal logic languages, such as description logic languages and first order logic languages. We have developed a web-related first order language (Web-PDDL) to represent ontologies and their mappings. We are continuously developing our inference engine, OntoEngine, which can do various inference tasks on the ontologies and the data described by the ontologies. The inference is the deduction process based on the relationships (rules) of concepts the ontologies specified. On the other hand, the interrelated data sets always show their relationships in a way which may not be represented as formal ontologies. We are going to build a machine learning tool, OntoLearner, especially use inductive logic programming to learn those relationships (rules) of the data instances from interrelated data resources or from those provided by the human through an user interface. The rules got by OntoLearner can be used for the inference tasks that OntoEngine needs to conduct, and OntoEngine also can test and optimize the generated rules and provide useful feedback to OntoLearner or even human users.

In a word, our goal is that OntoEngine and OntoLearner can be integrated together to provide a generic framework for different application areas, such as databases, data mining, the Semantic Web and biomedical informatics.

Faculty members: