Suranah writeup for Bunescu 2005

This is a review of the paper Bunescu_2005_a_shortest_path_dependency_kernel_for_relation_extraction by user:Suranah.

The paper attempts a different approach for relation extraction. They define a kernel which captures the shortest path between the two entities, in a dependency parse of the sentence. A higher dimensional vector maps each of these paths with respect to their categories or POS.

The kernel is then used to train a SVM. There is some degree of evaluation with different tree types, and different features. The results show some improvement. I find it intriguing that whether category classes can be learned, and if they can, whether the patterns can be used to train naive classifiers like in TextRunner.