penghts
commented
5 years ago First, you need to convert the text to a graph-of-words representation, and follow the node sequence selection, neighborhood assemble and neighborhood normalization steps to subgraphs-of-words. Second, you need to convert subgraphs-of-words to a 3-dimensional representation. Third, you can use the deep convolution neural networks to learn different-level of features. Fourth, for hierarchical multi-label classification tasks, you can even add recursive regularization of dependencies between labels into the traditional objective function.
Can you give a brief description on how to run the model on new dataset? Such as the order of running the scripts, how to build the graph, and so on.