Automatic training set generation

If you have an already existing set of user utterances from which you want to create an NLU you can use the Phrases clustering function that is available in the Advanced tools section

The main idea behind this technology is to categorize user phrases into a limited number of clusters which later on can be used as a training set for intent detection. The tool will result in generating a CSV that can be used as phrases that can be imported into NLU

Using this simple form you can upload user phrases in CSV (UTF-8) format. Remember that file should contain one column and each phrase should be in a separate row.

If you don't know the number of intents/clusters you can leave this field empty, however, the best results are achieved when there is a specific number of clusters. You can use this tool multiple times to find the best number.

By default clusters are not named - there will be only numbers that define their id. If you have OpenAI integration configured on your environment you can use it to create names for intents automatically. As a result you get a file that can be uploaded in the NLU module in the Phrases tab