TaxoPhrase: Exploring Knowledge Base via Joint Learning of Taxonomy and Topical Phrases

Knowledge bases restore many facts about the world. But due to the big size of knowledge bases, it is not easy to take a quick overview onto their restored knowledge. In favor of the taxonomy structure and the phrases in the content of entities, this paper proposes an exploratory tool TaxoPhrase on the knowledge base. TaxoPhrase (1) is a novel Markov Random Field based topic model to learn the taxonomy structure and topical phrases jointly; (2) extracts the topics over subcategories, entities, and phrases, and represents the extracted topics as the overview information for a given category in the knowledge base. The experiments on the example categories Mathematics, Chemistry, and Argentina in the English Wikipedia demonstrate that our proposed TaxoPhrase provides an effective tool to explore the knowledge base.