Document categorization using support vector machines

In this investigation a nu-SVC and string kernel-based categorizer of documents (undergraduate final projects) was developed. A corpus for training was generated from the compounded phrases (two-word phrases) and single words that were more representative of the selected areas (Electrical and Civilian Engineering), which were extracted from the document titles. The test set was made up with all the document titles of the undergraduate final projects between the years 1997 and 2006 (both years inclusive). The performance of the classifier, varying the parameters of the nu-SVC and the string kernel, was good after the tuning process. Results showed the great potential of the support vector machine in the text classification area.