A Discrete Kernel Approach to Support Vector Machine Learning in Language Independent Named Entity Recognition.
Bram Vanschoenwinkel · 2004
In this paper we discuss a discrete kernel approach to Support Vector Machine (SVM) learning to do Language Independent Named Entity Recognition (LINER). The kernel we use is called the polynomial overlap kernel (POK). It is derived from a distance function that has been succesfully used in memory-based learning for various natural language problems. The POK is a discrete function and it works on discrete examples represented by multi-valued string attributes. We will compare our approach to the bag-of-words approach making use of a continuous kernel function and we will argue that the POK and other related kernels can serve as a valid alternative to the bag-ofwords approach to do SVM learning in natural language settings.