Lernen und Interpretieren Strukturierter Dokumente - ein Qualitativer Ansatz

Technical interpretation is usually done on documents having a certain kind of structure. In the overall <br>layout of such documents components belonging together are grouped and hence geometrically <br>separated from other parts. The rules used for arranging components on <br>documents depend on the document's domain, the author's cultural background and many other things. <br>In order to extract the meaning of such document parts, the creation rules have to be known. <br>Therefore, a graph representation of structured documents as well as domains of such documents <br>has been developed, whereby the nodes stand for the components and the edges represent weighted <br>qualitative spatial relations among nodes. These relations are derived from Allen's qualitative <br>relations among time intervals. The representation of domains can be derived automatically from <br>sets of labeled documents. <br>From a learned models, consistent scenarios can be extracted and then be visualized. This makes it <br>more easy for the user to examine learned models. Models of domains can then be used to assign <br>labels to unseen documents by means of a heuristic search for inexact subgraph isomorphisms <br>between a model graph and a document graph.

Author dc:creator

• Walischewski, Hanno

Contributors dc:contributor

• Nebel, Bernhard

• machine learning
• qualitative spatial representation
• document layout interpretation

Repository record source_url

https://freidok.uni-freiburg.de/data/28

OAI identifier oai:identifier

oai:freidok.uni-freiburg.de:28