Nlao writeup of Bunescu 2006
This is a review of Bunescu_2006_subsequence_kernels_for_relation_extraction by user:Nlao.
Using dynamic programming to calculate subsequence kernel is a smart algorithm. However, the paper is lack of motivative example for the need of using subsequence.
The truth is that simpler full sequence kernel (by the same authors, Bunescu & Mooney, 2005) actually produces better result F1=52.5 (SP-CFG) vs. F1=47.7 (ERK). Not mentioning these connections is bad.