Bayesian Networks applied to Knowledge Base of a Never-Ending Learning System
Machine Learning, Never-Ending Learning, Bayesian Networks
The first never-ending learning system described in the literature is called NELL (Never-Ending Language Learning). The main objective of the NELL system is to learn to read the web, getting better each day, in order to store the acquired knowledge in a growing and never-ending knowledge base. To help the NELL system in this learning task, this project proposes the application of Bayesian networks to make inferences in the NELL knowledge base, as well as to identify new semantic relations that can be inserted in the knowledge base. A dataset was built from information on the semantic relations existing in the knowledge base of the NELL system. The results found show that Bayesian networks induced by the DMBC and K2 learning algorithms can represent existing relationships and suggest new relationships to extend the initial ontology of NELL. The inferences made also indicate that the Bayesian networks induced by both algorithms are capable of discovering new information to be inserted in the NELL knowledge base.