Text document clustering plays an important role in providing intuitive
vigation and browsing mechanisms by organizing large amounts of information
to a small number of meaningful clusters. Standard partitional or agglomerative
ustering methods efficiently compute results to this end.
wever, the bag of words representation used for these clustering methods is often
satisfactory as it ignores relationships between important terms that do not
-occur literally. Also, it is mostly left to the user to find out why a particular partitioning
s been achieved, because it is only specified extensionally. In order to
al with the two problems, we integrate background knowledge into the process of
ustering text documents.
rst, we preprocess the texts, enriching their representations by background knowledge
ovided in a core ontology — in our application Wordnet. Then, we cluster
e documents by a partitional algorithm. Our experimental evaluation on Reuters
wsfeeds compares clustering results with pre-categorizations of news. In the experiments,
provements of results by background knowledge compared to the baseline
n be shown for many interesting tasks.
cond, the clustering partitions the large number of documents to a relatively small
mber of clusters, which may then be analyzed by conceptual clustering. In our approach,
applied Formal Concept Analysis. Conceptual clustering techniques are
own to be too slow for directly clustering several hundreds of documents, but they
ve an intensional account of cluster results. They allow for a concise description
commonalities and distinctions of different clusters. With background knowledge
ey even find abstractions like “food” (vs. specializations like “beef” or “corn”).
us, in our approach, partitional clustering reduces first the size of the problem
ch that it becomes tractable for conceptual clustering, which then facilitates the
derstanding of the results.

@techreport{hotho03textclustering, author = {Hotho, Andreas and Staab, Steffen and Stumme, Gerd}, title = {Text Clustering Based on Background Knowledge}, type = {Technical Report }, year = {2003}, volume = {425}, url = {http://www.kde.cs.uni-kassel.de/stumme/papers/2003/hotho2003text.pdf}, keywords = {2003, analysis, background, clustering, concept, fca, formal, knowledge, myown, ontologies, semantic, text, web}, abstract = {Text document clustering plays an important role in providing intuitive navigation and browsing mechanisms by organizing large amounts of information into a small number of meaningful clusters. Standard partitional or agglomerative clustering methods efficiently compute results to this end. However, the bag of words representation used for these clustering methods is often unsatisfactory as it ignores relationships between important terms that do not co-occur literally. Also, it is mostly left to the user to find out why a particular partitioning has been achieved, because it is only specified extensionally. In order to deal with the two problems, we integrate background knowledge into the process of clustering text documents. First, we preprocess the texts, enriching their representations by background knowledge provided in a core ontology — in our application Wordnet. Then, we cluster the documents by a partitional algorithm. Our experimental evaluation on Reuters newsfeeds compares clustering results with pre-categorizations of news. In the experiments, improvements of results by background knowledge compared to the baseline can be shown for many interesting tasks. Second, the clustering partitions the large number of documents to a relatively small number of clusters, which may then be analyzed by conceptual clustering. In our approach, we applied Formal Concept Analysis. Conceptual clustering techniques are known to be too slow for directly clustering several hundreds of documents, but they give an intensional account of cluster results. They allow for a concise description of commonalities and distinctions of different clusters. With background knowledge they even find abstractions like “food” (vs. specializations like “beef” or “corn”). Thus, in our approach, partitional clustering reduces first the size of the problem such that it becomes tractable for conceptual clustering, which then facilitates the understanding of the results.} }

%0 = techreport %A = Hotho, Andreas and Staab, Steffen and Stumme, Gerd %B = } %C = %D = 2003 %I = %T = Text Clustering Based on Background Knowledge} %U = http://www.kde.cs.uni-kassel.de/stumme/papers/2003/hotho2003text.pdf

@inproceedings{stumme96attribute, author = {Stumme, Gerd}, title = {Attribute Exploration with Background Implications and Exceptions}, editor = {Bock, H.-H. and Polasek, W.}, booktitle = {Data Analysis and Information Systems. Statistical and Conceptual approaches. Proc. GfKl'95. Studies in Classification, Data Analysis, and Knowledge Organization 7}, publisher = {Springer}, address = {Heidelberg}, year = {1996}, pages = {457-469}, url = {http://www.kde.cs.uni-kassel.de/stumme/papers/1995/P1781-GfKl95.pdf}, keywords = {1996, acquisition, analysis, attribute, background, concept, exploration, fca, formal, implications, knowledge, lattices, myown} }

%0 = inproceedings %A = Stumme, Gerd %B = Data Analysis and Information Systems. Statistical and Conceptual approaches. Proc. GfKl'95. Studies in Classification, Data Analysis, and Knowledge Organization 7 %C = Heidelberg %D = 1996 %I = Springer %T = Attribute Exploration with Background Implications and Exceptions %U = http://www.kde.cs.uni-kassel.de/stumme/papers/1995/P1781-GfKl95.pdf