Structured knowledge about entities increasingly plays an important role in various web applications (e.g. search engines, recommendation sites, and social networks). Such knowledge is available in ontologies such as SUMO or OpenCyc, but such ontologies often have important limitations, which cannot be addressed using deductive reasoning alone.
One important issue is that the available information may be conflicting. For instance, the SUMO ontology claims that a “creek” is disjoint from “river” while Wikipedia claims that “creek” is of type “river”.
Another key limitation is that available knowledge is often incomplete. For example, the SUMO ontology encodes knowledge about summer olympics games such as Basketball, Football and Handball, but mentions nothing about Beach volleyball.
Clearly, web applications would benefit from inductive reasoning methods that are able to make plausible inferences, beyond what can be obtained using deduction, e.g. for automated knowledge base completion and for managing conflicts.
Our work aims to develop a number of methods for inductive reasoning with logic-based structured knowledge from the web (i.e. description logics or existential rules). At the centre of this work are vector spaces embeddings together with an statistical inference machinery that allows us to make plausible inferences in a principled way. In the following, we give an overview about our recent work.
Inductive Reasoning about Ontologies Using Conceptual Spaces
In this work, we developed a Bayesian method which is inspired by cognitive models of category based induction. In particular, given the knowledge that concepts C1,...,Ck all have some property P, our aim is to determine the degree to which we can plausibly conclude that some other concept C has property P. To this end, the method we proposed relies on a pre-learned vector space representation of the entities. From this embeddings, we can derive a number of interpretable features. The plausibility that C has property P is then obtained using a form of Bayesian inference over the resulting feature values.
We argue that this approach has several important advantages. First, since it is similar in spirit to cognitive models for category based induction and relies on interpretable representations, we can naturally generate intuitive explanations for all inference results. Second, the use of Bayesian inference supports a cautious form of inference, drawing conclusions only when it is warranted by the available evidence, and yields confidence scores in a principled way. Finally, unlike most existing methods, this method can be used for both TBox and ABox reasoning.
Zied Bouraoui, Shoaib Jameel and Steven Schockaert: Inductive Reasoning about Ontologies Using Conceptual Spaces. Proceedings of the thirty-first AAAI Conference on Artificial Intelligence (AAAI 2017). pdf