Exploiting semantic linkages among multiple sources for semantic information retrieval

Abstract

The vision of the Semantic Web is to build a global Web of machine-readable data to be consumed by intelligent applications. As the first step to make this vision come true, the initiative of linked open data has fostered many novel applications aimed at improving data accessibility in the public Web. Comparably, the enterprise environment is so different from the public Web that most potentially usable business information originates in an unstructured form (typically in free text), which poses a challenge for the adoption of semantic technologies in the enterprise environment. Considering that the business information in a company is highly specific and centred around a set of commonly used concepts, this paper describes a pilot study to migrate the concept of linked data into the development of a domain-specific application, i.e. the vehicle repair support system. The set of commonly used concepts, including the part name of a car and the phenomenon term on the car repairing, are employed to build the linkage between data and documents distributed among different sources, leading to the fusion of documents and data across source boundaries. Then, we describe the approaches of semantic information retrieval to consume these linkages for value creation for companies. The experiments on two real-world data sets show that the proposed approaches outperform the best baseline 6.3–10.8% and 6.4–11.1% in terms of top five and top 10 precisions, respectively. We believe that our pilot study can serve as an important reference for the development of similar semantic applications in an enterprise environment.

Keywords:

• enterprise application
• Semantic Web
• enterprise search
• linked data
• knowledge management

Acknowledgement

This first author contributed to this paper during his work at NEC Labs China.

Funding

This research was also supported by the National 973 project [grant no. 2011CB302505].