A hybrid semantic discovery approach to capture concepts, attributes and relationships

Building ontology from scratch need identify the basic concepts, and their attributes and relationships. In terms of information integration, draft concepts, attributes and relationships can be directly captured by processing schemas of data sources. In this context, we present a hybrid semantic discovery approach, which is an extension of our previous work[1][2] to capture concepts, attributes and relationships from relative schemas. The approach consists of two stages: column oriented matching phase and schema oriented matching phase. Column oriented matching phase focuses on catching a set of attributes of potential concepts by schema columns similarity computing using a composite matcher, and columns clustering using neural network matcher. Schema oriented matching phase categorizes schemas (regarded as potential draft concepts) into clusters or explicit concepts, and attaches relative attributes to the concept through high-dimensional sparse clustering process, which computes the relationship or semantic distance between potential concepts by comparing the corresponding attributes set of each potential concept. The explicit concepts with attributes discovered by our approach can be used as draft material or seeds for further ontology modeling.