Mining rules from knowledge graphs

This thesis explores the topics of knowledge graphs and mining association rules from them. Knowledge graphs are large data structures which store linked data that describe the relations between entities in a semantic way. The massive quantities of data stored in those graphs are a great source of material for data mining analysis. One of the most explainable data mining methods that focuses on finding relationships and patterns is association rule mining. Biochemistry is one of the most progressive fields in terms of publishing linked data. One example of this effort is KG-Microbe which is a knowledge graph that connects data concerning microorganisms and their features and functions. The experiment which is described in this thesis uses a special tool called RDFRules for mining association rules from this knowledge graph. Other experiments have been conducted on the data from this knowledge graph in a flattened format and using less explainable machine learning methods. This work describes the process of creating an association-rule-based classifier to make predictions about the cultivation medium of a microbe. While the accuracy of this classifier is not on par with a referential CatBoost classifier, the results are highly interpretable and can be used for further analysis of the new-found patterns in the graph.