Relation extraction with tree kernel for Indonesian sentences

This paper propose of a study about kernels method for relation extraction in natural language tasks. Our study based on relation extraction using Indonesian parse tree kernel approach such as define subtree and subset tree, establish word dependency pattern and modified parse tree algorithm for testing accuracy of the dependency tree. We aim to modified the algorithm that was created by Moschitti in[2] for the evaluation of the sub tree (ST) and subset tree (SST) kernel using Indonesian sentences. The concept of Indonesian tree kernel is when production associated with n1 and n2 are different, we can avoid to evaluate delta (n1,n2) since it is 0. For experiment, we extract relation from various offline resources, with domain including academic papers in the fielsd of language practicing and science. By using 250 sentence, the highest accuracy score for node pair set `Role_Staff ?Role owner' achieved 89,2% for ST and 86,5% for SST.