Language sensitive text classification

It is a traditional belief that in order to scale-up to more effective retrieval and access methods modern Information Retrieval has to consider more the text content. The modalities and techniques to fit this objectives are still under discussion. More empirical evidence is required to determine the suitable linguistic levels for modeling each IR subtask (e.g. information zoning, parsing, feature selection for indexing,...) and the corresponding use of this information. In this paper an original classification model sensitive to document syntactic information and characterized by a novel inference method is described. Extensive experimental evidence has been derived on real test data and also from well-established academic test sets. The results show that a significant improvement can be derived using the proposed inference model. Also the role of linguistic preprocessing seems to provide positive effects on the performance. POS tagging and recognition of Proper Nouns received a specific experimental attention and provided significant effects on measured accuracy.