From Semi-automated to Automated Methods of Ontology Learning from Twitter Data

This paper presents four different mechanisms for ontology learning from Twitter data. The learning process involves the identification of entities and relations from a specified Twitter data set, which is then used to produce an ontology. The initial two methods considered, the Stanford and GATE based ontology learning frameworks, are both semi-automated methods for identifying the relations in the desired ontology. Although the two frameworks effectively create an ontology supported knowledge resource, the frameworks feature a particular disadvantage; the time-consuming and cumbersome task of manually annotating a relation extraction training data sets. As a result two other ontology learning frameworks are proposed, one using regular expressions which reduces the required resource, and one that combines Shortest Path Dependency parsing and Word Mover’s Distance to fully automate the process of creating relation extraction training data. All four are analysed and discussed in this paper.