一個知識分類與搜尋相關資訊的架構

A Framework for Knowledge Classification and Related-Information Search

黃美珠(Mei-Chu Huang)

指導教授 ： 夏延德

中原大學/工學院/資訊工程研究所/碩士(2001年)

https://doi.org/10.6840/cycu200100331

摘要

目前搜尋資訊的方式，大部分都是以關鍵字搜尋為主，再配合幾個大分類。這種搜尋方式，只比對符合的字串，通常沒有對資訊的內容做詳細分析，因此搜尋結果往往會出現很多不相關的資訊。要使搜尋到的資訊是內容真正相關的資訊，首先，分類的架構必須從”知識”的角度出發，把domain knowledge含入，因此每個分類架構就是一顆代表著某個domain的知識樹。再來，必須對資訊內容做詳細分析後，從知識樹上選擇足以代表資訊特徵的知識類別。由於在分類建構過程中，已經依據知識樹對資訊的內容做詳細分析，因此資訊間的相關性，便由知識類別來主導。

在本篇論文中，以知識樹為分類架構的基礎，並藉由多重角度的知識分類，從各個不同層面將資訊內容做詳細分析後建構成知識庫，因此資訊所屬的知識類別將成為搜尋相關資訊的主要依據，且搜尋到的資訊是內容真正相關。同時並設計出一套規則及計算方式，將資訊間的關聯程度做量化，所以資訊間將依據知識類別，產生不同程度的關聯，可進一步做為不同層級的知識應用。最後以二個實作的範例─科技新聞與西洋繪畫，測試搜尋相關資訊的結果及資訊與所屬知識類別間的相關程度。

關鍵字

關聯表 ； 多重角度 ； 知識模板 ； 知識分類 ； 知識樹

並列摘要

Today, most content-based search methods use keywords to search the information-items. Sometimes general classifications are also considered in the search. These search methods greatly depend on the use of keywords. They seldom do reasonable analysis of the real contents of the information-items. As a result, some of the information-items found will turn out to be irrelevant.

In order that all information-items found are relevant, it is advisable that we first classify these information-items appropriately. This classification is necessarily knowledge-based. One way of doing this is to use our domain knowledge to construct a "knowledge tree", and then to categorize each information-item under its related categories. This in turn means that we need to analyze the contents of the information-items. Once we do so, the relatedness between any two information-items will be implicitly represented by the relatedness between their "immediate enclosing categories."

In this thesis, we first turn our (domain-dependent) classification knowledge into a knowledge tree, and then we use this knowledge tree as a basis for classifying the information-items. Any information-item can be classified into multiple categories. This implements the idea of using multiple perspectives to view things. Since the relatedness of the information-items is implicitly expressed by the structure of the knowledge tree, information-items in the same "immediate closing category" are related by definition. To further calculate the relatedness between any two information-items, we also designed a set of computation rules. This allows us to quantify relatedness.

To show how our approach works, we did two case studies. One is about the classification of (a small domain of) high tech news-items. The other is about the classification of museum items. Both studies demonstrate how related information-items can be found and suggested to the viewer of the current information item.

並列關鍵字

related table ； knowledge classification ； knowledge tree ； multiple viewpoint ； knowledge template

參考文獻

[1]Jacobs, P.S. Using statistical methods to improve knowledge-based news categorization. IEEE Expert [see also IEEE Intelligent Systems], Volume: 8, Issue: 2, April 1993, Page(s): 13 - 23.

[3]Blasius, K.-H.; Grawemeyer, B.; John, I.; Kuhn, N. Knowledge-based document analysis. Document Analysis and Recognition, 1997. Proceedings of the Fourth International Conference on, Volume: 2, 1997, Page(s): 728 - 731.

[5]Abu-Hanna, A.; Jansweijer, W. Modeling domain knowledge using explicit conceptualization. IEEE Expert [see also IEEE Intelligent Systems], Volume: 9, Issue: 5, Oct. 1994, Page(s): 53 - 64.

[6]B. Chandrasekaran, Todd R. Johnson and Jack W. Smith. Task-structure analysis for knowledge modeling. Commun. ACM, Vol. 35, No. 9 (Sep. 1992), Page(s): 124 - 137.

[7]P. S. Jacobs and Lisa F. Rau. SCISOR: extracting information from on-line news. Commun. ACM, Vol. 33, No. 11 (Nov. 1990), Page(s): 88 - 97.