Analysis of relation types in semantic network used for text classification

Abstract

Contemporary information systems contain lots of textual information. One of important kinds of textual information processing is text classification. Semantic network is a model which can be used to resolve different tasks including text classification. There are other models which can resolve the same task, and some of them show relatively good results, but using semantic networks has such advantages as human readability and analyzing actual semantic relations between words. Semantic networks have different set of relation types depending on the network purpose. This article is devoted to looking for a set of relation types to be leveraged in semantic network created for text classification. The analysis is performed by generating semantic networks for Russian-language texts leveraging different sets of relation types. Generated networks are used for text classification. Texts were taken from books on several technical disciplines. Proposed algorithms can be used to perform text classification when performing such tasks as dividing electronic messages on categories, spam filtering, text topic recognition and other.