I'm currently working on the classification with massive amount of data. Similar to the kaggle one. Data input consist of features and multiple labels that can be hierarchically aligned.
At first I flattened the data and tried to learn multi-label classifier. That involved both, methods that are specially designed for it (like multi-label kNN) and One-vs-All methods. However these approaches didn't yield much result and were really complex.
Afterwards I found structured SVM library. Description says, that this library actually minds the structure in the data. However due to its complexity and lack of examples, I didn't have enough time to try it myself.
I was thinking maybe someone could shed some light to other hierarchical classification methods out there (in R, Python, C or Java), especially the guys that were tackling that problem on kaggle. What's your approach to this?