A Semantic Integration Approach for Building Knowledge Graphs On-Demand

Abstract

Information about the same entity may be spread across several Web data sources, e.g., people on the social networks (Social Web), product descriptions on e-commerce sites (Deep Web) or in public Knowledge Graphs (Web of data). The problem of integrating entities from heterogeneous Web data sources on-demand is still a challenge. Existing approaches propose expensive Extraction Transformation Loading (ETL) processes and rely on syntactic comparison of entity properties, leaving aside the semantics encoded in the data. We devise FuhSen, an integration approach that exploits search capabilities of Web data sources and semantics encoded in the data. FuhSen generates Knowledge Graphs in response to keyword-based queries. Resulting Knowledge Graphs describe the semantics of the integrated entities, as well as the relationships among these entities. FuhSen approach utilizes an ontology to describe the Web data sources in terms of content and search capabilities, and exploits this knowledge to select the sources relevant for answering a keyword-based query on-demand. The results of various empirical studies of the effectiveness of FuhSen suggest that the proposed integration technique is able to accurately integrate data from heterogeneous Web data sources into a Knowledge Graph.