Hierarchical pattern matching for knowledge based news categorization
Luca Gilardoni, Paola Prunotto, Gianluigi Rocca · Multimedia Information Retrieval · 1994
In this paper we describe how we have combined the use of pattern matching with the exploitation of relation links between objects of a knowledge base, in order to perform a first-level broad categorization (shallow categorization) of natural language texts (agency news). This work has been done in the context of the COBALT system, a text categorization shell which we are developing in the framework of the CEC funded COBALT project (LRE, Linguistic Research and Engineering project n° 61-011). In the COBALT system, concepts are defined as frame like objects organized in a hierarchy for which properties, relations and patterns may be defined. The patterns associated to the concepts' definitions are aimed at identifying the concepts within the text through pattern matching whilst the properties and relations may be used in order to perform categorization according to their specific values in the knowledge base. An inheritance mechanism has been implemented for concept matching inheritance and a rule language has been defined which allows to exploit this inheritance mechanism besides allowing to navigate within the knowledge base structure.