Hi There,
as part of my ongoing attempt to do non-IoT things with Node-RED, this time automatic text classification:
What this gif shows is a drag&drop feature combined with a HuggingFace text2text model for doing automatic text classification.
The idea is to drag text onto Node-RED and have NR create a node containing the text plus topic nodes that classify the text. As more text is dropped and included, the topic nodes cluster together creating relationships between individual snippets of text. Existing topic nodes are connected to new texts on the right side of the topic node while new topic nodes are connected on the left side - hence a pathway is created as more text is dragged onto Node-RED.
At the moment the classification only works with texts but the same classification could be done with images and an AI model that describes the image.
The purpose behind all this is to create mind maps in Node-RED. What this could look like is demonstrated by a beta version which also supports drag&drop but not AI - the text classification is faked using word frequencies. The resulting flows can be exported and have the same format as Node-RED flow jsons, so there is a 1-2-1 compatibility with existing flows.
What this means is that my personal mind map also has a web server for serving up nodes from my mind map. It also contains a file upload so that large content is store to disk and not into the flow.json file.
