I would like to do some text classification work with Knime.
I am already familiar with the general flow of text classification on Knime, and I would like to use Knime to handle a hierarchical classification task, where the output is a hierarchical structure such as “Dresden“-”German“-”Europe”. But unfortunately, I haven’t found any workflow for hierarchical classification yet, and I would like to know if it is possible to implement global classification in Knime in addition to Flat and lokal Classification.
Sadly, I do not know of a workflow that is designed to demonstrate working with hierarchical text structures.
However, I can point you towards two possibly helpful resources:
The newly released self-paced course on text processing. You probably know most of the content already, but it could be a nice refresher that might even present one or two new tricks
A blog article showing how to do text classification using Google’s BERT neural network architecture.
I’ll update you if I find something more related to your use case.
I suspect that what you will need is to bring in some extra data to allow your workflow to understand that Dresden is a city in the country Germany which is in the continent of Europe. I’d suggest looking at the Semantic Web extension for the SPARQL tools to allow you to query wikidata for this extra information. There is a guide on querying wikidata with knime here and some example queries for wikidata to try and modify here
Perhaps you can try out our new Redfield NLP Nodes extension. There is no exact hierarchical classification nodes, however it has multi-label classification with BERT functionality. And the following example demonstrates training the model for text classification to assign the labels with simple 2-level hierarchy. Hope it will be useful for you.
