Metadata Based Matching of Documents and User Profiles
Eerika Savia, T. Kurki, Sami Jokela · 1998
The growing amount of new information has created a need for information filtering. Filtering is usually done on the textual content of the documents. Recent developments in the field of metadata suggest that filtering could be done according to only metadata (description of the actual document content). In information filtering documents are matched against user interest profiles. This is based on some measure for similarity or distance between a representation of documents and user profiles. Both the representation and the distance measure should make comparisons meaningful. A common problem is caused by closely related concepts that are considered independent in the representation model. Furthermore, the matching should not be considered symmetric, since documents may cover some area of interest very well and should be matched against parts of the user profile instead of the whole profile. In this paper we suggest a hierachical representation for describing documents and user profiles that attempts to model the related concepts. Our model includes an asymmetric distance measure that can also detect documents that cover some subtopic of an interest profile.