Web Information Extraction for the Creation of Metadata in Semantic Web

Web Information Extraction for the Creation of Metadata in Semantic Web Ching-Long Yeh and Yu-Chih Su Department of Computer Science and Engineering Tatung University Taipei, Taiwan chingye@cse.ttu.edu.tw g9206033@ms2.ttu.edu.tw Abstract In this paper, we develop an automatic metadata creation system using the information extraction technology for the Semantic Web. The information extraction system consists of preparation part that takes written text as the input and produces the POS tags for the words in the sentences. Then we employ finite state machine technology to extract the units from the tagged sequences, including complex words, basic phrases and domain events. We use the components of an NLP software architecture, GATE, as the processing engine and support all required language resources for the engine. We have carried out an experiment on Chinese financial news. It shows promising precision rate while it need further investigation on the recall part. We describe the implementation of storing the extracted result in RDF to an RDF server and show the service interface for accessing the content. 1. Introduction Semantic Web is an emerging technology working by building a metadata layer upon the current web and using the metadata description language to describe the resources on the WWW [1]. The metadata layer is a structured information layer, that is, the information model of RDF [2]. The service programs built upon the metadata layer can provide an efficient way for users to find the information they need according to the concepts. Furthermore, the metadata layer supports agents to provide the function of service automation for users, called Semantic Web Service [3]. The semantic metadata layer can be constructed either in a manual way using annotation tools, such as Annotea [4] or automatically using the natural language processing technology [5]. The former approach is good at higher precision rate while suffered in efficiency. The latter approach can be used to sort out the problem of the other; however, it needs very high cost to achieve similar result using the manual way. Information extraction is a low-cost approach to natural language processing using the finite-state automata technology to extract specific noun sets and information matching specific syntax and semantic templates [6]. In this paper we employ the finite-state automata technology to extract information from the resources on the WWW to serve as the describing information of the resources, that is, the metadata. The extracted metadata is then used as a part in the Semantic Web. Ontology, the semantic schema used to build metadata layer upon semantic web, contains the concept hierarchy of specific domain knowledge and the detail features of all concepts type. In this paper we build the templates for information extraction based on the feature structure of concepts type of ontology. A typical information extraction system, such as FASTUS of SRI [7], ANNIE of the University of Sheffield [8], is constructed by using a cascade of finite-state grammars. Based on such mechanism, we can identify the units of words, complex words, basic phrase group, co-reference resolution, identifying event structures etc., …

Content truncated to 500 words