[互联网]软件前沿_知识工程new.ppt

* * * * changing the world by spreading the knowledge of innovators * * * * By extracting profile information, we take homepages of researchers as data sources. Here is a typical example. The information we want can be divided into three categories, first contact information, like office, post address, email, phone or fax. The second category, basic information, here like the researcher’s name, his personal homepage url, the affiliation and the job title. The last, we call it educational history, which include the university where he graduated, in which major he studied and the exact graduation date. * * At first, let us see what are the main problems in this area. As we know, resources are in various formats and have different characteristics in the content description. That means, at first, almost all types information exist on the Web. For example, the data we want may be located in natural language description, in email and in semi-structured docuements. Furthermore, the data may be represented in web pages, formatted tables, and even unformatted table. e. Secondly, it means that the date embedded in the resources have different characteristics. Some are organized in template format and some are non template . So, different characteristic resources require different semantic annotation methods. * In this work, we proposed a ontology matching framework with strategy selection. At first, We implemented many matching strategies by using different kinds of information defined in ontology. Then we defined two similarity factors to characterize the similarity between two ontologies. Then we use these factors to select different ontological information and different matching strategies for different tasks. 我们提出的方法参加了2008年OAEI 国际本体映射比赛，在参赛的5个任务中，有四项都取得取得了前三名的好成绩。这是其中三项的比赛结果。 * * Forward chaining: 预先推理出结果 RETE-Algorithm Backward chaining:根据目标进行推理 Xcerpt * * * Intermediate approaches: firstly convert one data model to the other, and then reuse certain schema matchin