Exploiting co-occurrence of low frequent terms in patents

Abstract

This paper investigates the role of co-occurrence of low frequent terms in patent classification. A comparison is made between indexing, weighting single term features and multi-term features based on low frequent terms. Three datasets are used for experimentation. An increase of almost 21 percent in classification accuracy is observed through experimentation when multi-term features based on low frequent terms in patents are considered as compared to when all word types are considered.