Ontology-based data integration for supporting big bridge data analytics

The deterioration of bridges is dependent on complex interactions of multiple factors. Existing research efforts have focused on predicting bridge deterioration using indicators, which are limited in capturing the many deterioration factors and the interactions between them. On the other hand, a large amount of bridge data is being generated, which opens opportunities to big bridge data analytics for improved bridge deterioration prediction. Such bridge data include: (1) National Bridge Inventory (NBI) and National Bridge Elements (NBE) data, (2) traffic, weather, climate, and natural hazard data, and (3) data from bridge inspection reports. There is, thus, a need for data integration methods that are able to integrate bridge data from multiple sources and in heterogeneous formats. To address this need, this paper proposes an ontology-based data integration methodology. Ontology aims to facilitate the integration based on content and domain-specific meaning. The proposed methodology includes two primary components: (1) ontology-based data linking: identifying the links among data from different sources, and (2) ontology-based data fusion: resolving conflicts between the linked data and then fusing the conflict-resolved linked data. This paper focuses on presenting the proposed ontology-based data linking methodology and its experimental results. Data linking is defined as a multi-class classification problem - classifying data links into multiple types, including "is-type-of, "is-supertype-of, "is-part-of, "is-parent-of, "is-related-to", "isequivalent-to", and "has-no-match". In developing the methodology, several comparison functions (for comparing the similarities between attribute values) and machine learning algorithms (for the classification of data links) were implemented and evaluated based on accuracy. The experimental results show that the proposed data linking methodology achieved an accuracy of 98.7%.