Template-based Information Extraction from Tree-structured HTML Documents

• Main Authors: Yih, Wen-tau, 易文韜
• Other Authors: Jane Yung-jen Hsu
• Format: Others
• Language: zh-TW
• Published: 1997
• Online Access: http://ndltd.ncl.edu.tw/handle/29386387918552988914

碩士 === 國立臺灣大學 === 資訊工程學系研究所 === 85 === This thesis proposes a novel approach of information extraction by identifying structural components in on-line web documents. The brief description of this approach can be introduced as follows. First, documents can be modeled using the three ingredients: format, content, and structure. Format is the visualized view of a document. Content is the actual data that a document has. Structure is the logical organization of a document. Structural components can be used as extraction targets. If a mapping relationship has been identified, the corresponding content will then be extracted. With the he lp of content and format, the procedure of finding the mapping relationship is facilitated by a template, which is us ed to simulate the structure of a document. When applying the document model and the concept of template matching to HTML documents, both documents and templates can be represented in tree-structures. The matchin g problem between templates and documents can then be transf ormed into an approximate tree matching problem.After a mapping r elationship has been built, the content that belongs to the target structural component will be the result returned by the s ystem. Experiments have shown the feasi bility of this template-based information extraction approac h. In addition, the systematic procedure of building templa tes also implies the possibility of generation of templates by the means of machine learning.