الفهرس 
Background and Related WorkOnly 14 pages are availabe for public view
 |  | from | 147 |  |  |
| | | from | 147 | | |
Abstract
Abstract
Property Graph (PGs) databases are emerging as efficient graph stores with flexible schemata. This raises the need to have a unified view over heterogenous data produced from these stores. Ontology based Data Access (OBDA) has become the most dominant approach to integrate heterogeneous data sources by providing a unified conceptual view (ontology) over them. The corner stone of any OBDA system is to define mappings between the data source and the target (domain) ontology.
Manual mapping generation is time consuming and requires great efforts. This thesis proposes ProGOMap (Property Graph to Ontology Mapper) system that automatically generates mappings from property graphs to a domain ontology. The main building blocks of the proposed system include generating a putative ontology from PG, aligning the putative ontology to an existing domain ontology, and generating mappings from alignment results.
Additionally, we propose an Ontology Based Data Access (OBDA) approach to query NoSQL property graph databases using SPARQL language. This allows users to access diverse data sources with the same query through a unified conceptual view (ontology). The proposed OBDA system goes through an offline phase to map the property graph data model to a unified ontology model. The online phase starts upon SPARQL query submission, and its building blocks include SPARQL query unfolding, query syntax tree optimization, and property graph Cypher query construction and execution.
In order to assess the proposed PG-to-Ontology mapping generation approach, experiments were conducted on eight data sets with different scenarios to evaluate the accuracy of the generated mappings. Evaluation metrics include precision, recall, and F-measure scores. The experimental results achieved mapping accuracy up to 97% and 81% when addressing PG-to-ontology terminological and structural heterogeneities, respectively.
Moreover, two datasets, in the commercial domain, with different sizes and schema are employed to perform a reliable evaluation of the proposed SPARQL-to-Cypher translation approach. Evaluation metrics include measuring the query unfolding time, the query translation overhead, the overall response time, and the impact of query optimization. Additionally, the proposed approach is compared against materializing the property graph over traditional triple stores. Experimental results show that the translation overhead remains negligible as long as the SPARQL triple patterns do not share common RDF terms.