關聯式資料庫至知識本體之對映: 一套結合豐富語意與對映一致性的方法

Mapping Relational Databases to Ontologies: An Approach Combining Semantic Enrichment and Mapping Consistency

劉俊宏(Jun-Hong Liu)

指導教授 ： 蔡益坤

國立臺灣大學/管理學院/資訊管理學研究所/碩士(2008年)

https://doi.org/10.6342/NTU.2008.00999

摘要

語意網的核心挑戰之一就是將現有的大量資訊轉變成由知識本體語言所定義的知識本體，這些現有的大量資訊主要是由網路上的資訊內容所構成，因此，為了實現語意網的目的，將這些網路上的資訊內容對映至知識本體是有必要的。既然網路上擁有超過數億的網頁，而這些網頁上的資訊內容大多數是被儲存在關聯式資料庫因而很難被搜尋引擎所找到(俗稱“深網”)。因此，為了讓語意網可以取用這些資訊內容，一個有效的方法就是將深網底下的關聯式資料庫對映至某特定領域中一個已有的知識本體。
在這篇論文中，我們設計出一套半自動化方法直接將一個關聯式資料庫對映到某特定領域中一個已有的知識本體，這個方法採取在資料採礦中的群集分析概念去替每個表格找出它們的匹對類別群組，其中在類別群組中的每個類別都會滿足對映一至性亦即對映的結果不應該違背在關聯式資料庫所表達的事實。我們將此方法分成兩階段。第一階段主要是利用語意上的資訊將關聯式資料庫中的外鍵對映到知識本體中的物件屬性以獲得一些表格的匹對類別群組特徵，有別於其它方法，我們在對映時，不只考量了表格之間的一般化/特殊化關係也考量了隱含表達在關聯式資料庫中的反關係以建構出一個隱含匹對等級。第二階段使用外鍵的對映結果作為特徵線索以找出一些表格的匹對類別群組，然後再利用這些表格的匹對類別群組去找到其它表格的匹對類別群組，最後將每個表格對映至本身的匹對類別群組。一個雛形化系統顯示出利用此方法執行一些現實世界中的樣本對映有良好的績效。

關鍵字

語意網 ； 知識本體語言 ； 深網 ； 對映 ； 關聯式資料庫 ； 群集分析 ； 對映一致性

並列摘要

One of the core challenges of the Semantic Web is to transform mass existing information to OWL ontologies. This mass existing information is mainly composed of the contents on the Web. Therefore, to realize the Semantic Web, it will be necessary to map the Web contents to OWL ontologies. Manually mapping the Web contents to OWL ontologies is impractical since the Web has well over billions of Web pages and most of the contents of Web pages is stored in the relational databases and hence hard to be found by search engines (so-called the ``emph{deep Web}"). Hence, to make these contents available for the Semantic Web, an effective way is to map the relational databases underlying the deep Web to domain-related OWL ontologies.
In this thesis, we propose a semi-automatic approach for directly mapping relational databases to OWL ontologies. This approach takes the concept of cluster analysis in data mining to find the matching classes group (MCG) for every table, where every found class of MCG will satisfy the mapping consistency stating that the mapping results should not violate the fact expressed in the relational database. We divide our approach into two phases. The first phase primarily uses the semantic information to map the foreign keys of a relational database to object properties of OWL ontologies to get the features of MCGs for tables. Different from other approaches, we not only take into consideration the Generalization/Specialization relationship between tables but also take advantage of the inverse relationship between tables which are implicitly expressed in a relational database to construct an implicit matching level between foreigns keys and object properties. The second phase uses the mapping results of the foreign keys as clues of features to find MCGs for some tables and then takes the MCGs of these tables to find MCGs for other tables. Finally, every table is mapped to its MCG. A prototype system demonstrates that our approach performs well on several domain samples from the real world.

並列關鍵字

Semantic Web ； OWL ； Deep Web ； Mapping ； Relational Databases ； Cluster Analysis ； Mapping Consistency

參考文獻

[1] Y. An, A. Borgida, and J. Mylopoulos. Inferring complex semantic mappings between relational tables and ontologies from simple correspondences. International Semantic Web Conference., pages 6-20, 2005.

[2] I. Astrova. Reverse engineering of relational databases to ontologies. Proceedings of 1st European Semantic Web Symposium, Heraklion, Crete, Greece, LNCS, 3053:327-341, 2004.

[8] S. Benkley, J. Fandozzi, E. Housman, and G. Woodhouse. Data element tool-based analysis (delta). MITRE Technical Report MTR, 1995.

[10] M. K. Bergman. The Deep Web: Surfacing Hidden Value, 2000.

[12] E.F. Codd. A relational model of data for large shared data banks. Communications of the ACM, 13(6):377-387, 1970.