Using Positional Suffix Trees to Perform Agile Tree Kernel Calculation

Tree kernels have been used as an efficient solution for many tasks, but are difficult to calculate. To address this problem, in this paper we introduce the Positional Suffix Trees: a novel data structure devised to store tree structures, as well as the MFTK and EFTK algorithms, which use them to estimate Subtree and Subspace Tree Kernels. Results show that the Positional Suffix Tree can store large amounts of trees in a scalable fashion, and that our algorithms are up to 22 times faster than the state-ofthe-art approach.