Ontology Learning Approach Based on Analysis of the Context and Metadata of a Weakly Structured Content

摘要

This article describes ontology learning approach based on the analysis of metadata and the context of weakly structured content. Today, there is a paradigm shift in ontological engineering. It consists of the transition from manual to automatic or semi-automatic design. This approach is called ontology learning. When an author creates a document, one holds in one's head a model of a certain subject area. Then, analyzing the document, it is possible to restore the model of this subject area. This process is called reverse engineering. Current articles describe ontology learning approaches based on content analysis. We propose to use not only the content, but, if it is possible, its metadata and the context for ontology learning purposes. As the main results of the work, we can introduce the model for the joint presentation of content and its metadata in a content management system. To extract the terms, the ensemble method was used, combining the algorithms for extracting terms both with and without contrast corpus. Metadata was used to expand candidates attribute space. In addition, methods for constructing taxonomic relations based on the vector representation of words and non-taxonomic relations by analyzing universal dependencies are described.