METHODS FOR TRANSFORMATIONS RELATIONAL DATABASE TO ONTOLOGY

• Main Author: Paramita Mayadewi, Ra
• Format: Dissertations
• Language: Indonesia
• Online Access: https://digilib.itb.ac.id/gdl/view/75276
• Institution: Institut Teknologi Bandung

The use of ontologies in semantic computing has led to the emergence of research to explore methods of developing new ontology models. Building an ontology is an engineering activity, and there are two main approaches to building it: building from scratch (manually) or using an ontology learning approach. Formally, using a learning approach, ontologies can be constructed from various sources of information, including structured sources, such as relational databases; semi-structured sources, such as dictionaries, HTML or XML files or unstructured sources, such as web pages—most research topics in ontology learning focus on the extraction of knowledge stored in relational databases. Many studies have proposed converting relational databases into ontologies to explore knowledge from relational databases. Despite the significant progress made over the last few years and the many proposed approaches, many issues have not been adequately addressed, especially involving database instances to discover implicit knowledge from the source database. The research is about learning ontology from relational databases and aims to discover hidden knowledge in relational databases. The focus of the research is: First, the identification of symmetrical and transitive relationships in relational databases. Second, entity identification is based on attribute matching. Third, identify the property chain patterns in relational databases. Identifying symmetrical and transitive relationships can assist in deducing facts to be stated in the knowledge base and enriching the resulting ontologies. The research that has been done proposes a new method to identify symmetrical and transitive patterns. The proposed approach to identifying specific patterns of symmetric and transitive relationships is based on the patterns formed between primary and foreign keys. It is limited to unary relations and relationship tables. The proposed approach can identify symmetrical and transitive relationship patterns in unary and relationship tables. The process of identifying entities through attribute matching is part of the process of transforming a relational database into an ontology. It aims to reduce the duplication of data values in the ontology and enrich the resulting ontology. The proposed approach is data oriented in the database and does not check solutions that depend on external knowledge of the data. The proposed method uses Jaccard Similarity to identify attributes that are thought to have similar meanings so that they can be used as classes in ontology. Experimental studies are carried out by calculating the precision, recall and F1-score value. The results of an experimental study using three different thresholds (3%, 5% and 7%) show an average value of precision, recall and F1-score above 75%. The purpose of identifying ObjectPropertyChain is to discover new facts or knowledge hidden in relational databases. The ObjectPropertyChain identification approach focuses on ObjectPropertyChain of length two. The approach taken uses a graph representation which is implemented using a hashmap. The matching process is carried out by matching the second value of the attribute pair that appears as the first value of the attribute pair in a different table. The attribute pairs resulting from the chain property identification process will be compared with the data in the database. Each table or relation that satisfies the pattern identified will be proposed as an axiom that can be added to existing object properties or added, in fact, new object properties in the ontology. Experiments on several databases resulted in a precision value of 88.9%. The results of the transformation process are tested using information and query preservation. The validation is done by query experiments using SQL and SPARQL. The results of both queries are compared to see whether the approach proposed is information preservation or query preservation.