The GENIE System: Classifying Documents by Combining Mixed-Techniques

摘要

Today, the automatic text classification is still an open problem and its implementation in companies and organizations with large volumes of data in text format is not a trivial matter. To achieve optimum results many parameters come into play, such as the language, the context, the level of knowledge of the issues discussed, the format of the documents, or the type of language that has been used in the documents to be classified. In this paper we describe a multi-language rule-based pipeline system, called GENIE, used for automatic document categorisation. We have used several business corpora in order to test the real capabilities of our proposal, and we have studied the results of applying different stages of the pipeline over the same data to test the influence of each step in the categorization process. The results obtained by this system are very promising, and in fact, the GENIE system is already being used on real production environments with very good results.