| Summary: | 碩士 === 國立臺灣大學 === 資訊管理學研究所 === 96 === Web Services provide a standard means of interoperation between different software applications running on a variety of platforms and frameworks. Because of the rapid development of Web services,the scale of Web Services becomes larger. As a result, finding thesuitable Web Services is not a trivial task. Currently, the discovery mechanism supported by UDDI is not powerful enough for
automatic discovery. Since the mechanism is keyword-based search,the users have to type in a precise term to find the services thatthey want. Besides, it is necessary for people to read the returned description. Too many manual processes lack efficiency. The Semantic Web provides a common framework that allows data to be shared and reused across applications. Ontology is the basis of semantic annotation on Semantic Web. As a result, we can benefit from associating information with ontologies.
In this thesis, we propose a semi-automatic approach for mapping Web tables to ontologies. Although the collection of Web Services are getting larger, there are still many other applications working on the Web. Through our mapping approach, we can integrate abundant information on Web pages into our ontology. The purpose of our approach is to directly import and map data / information from a Web
page to an domain ontology. In this way, we have to firstly generate information from Web pages and then annotate the generated results. Accordingly, we can divide our approach into two main tasks. The first is ``Information Extraction", which aims to extract information from Web pages. The second is ``Semantic Annotation", which aims to generate the mapping between Web pages and ontologies.
In the real world, information on Web pages could be exploited in many forms. It is nearly impossible to find a way to solve all kinds of “Information Extraction” problems. Therefore, we narrow down the scope of ``Information Extraction" task to Web tables. Tables, no matter they are on Web pages, text files, or in database are used to present certain type of information to their viewers in a formatted way. We apply the observed heuristics to help us extract and analyze tables. In the ``Semantic Annotation" phase, Natural Language Processing tools are adopted to help us interpret and annotate the tables. Our framework therefore can directly store and
consolidate the corresponding instances and relationships in the ontology. We implemented a prototype system ``Table Mapper" to demonstrate our approach. From the execution results, our system performs well in several different Web pages.
|
|---|
