Continuous improvements on your dataset

The most common needs when using your dataset to classify real data are:
  • Analyze how your dataset is performing
  • Improve your dataset with real data

BotHub makes available a very handy tool to achieve both tasks, the Inbox 📥. It's a page that lists all the sentences that were predicted with corresponding result, that you can filter by Intent, Language or Version of your dataset. Let me show you how to use it:

  1. Open your dataset page and expand the sidebar to see Inbox option
  2. Click on the sentence you want to see details about. You will see the intents and entities identified for each
  3. If you can to see more details about each prediction, you can just click in Raw or Debug buttons at the right side
  4. At the end you can even add them to your training or testing data, just by selecting one or many. After that, just click on the icons on the top of the list and select one of the intents you want to add
  5. Now you can just train your dataset and now it's smarter than before with real data. If you want to filter the sentences you have a couple of options
  6. Now you can just explore the real predictions of your dataset and use it to improve!

How did we do?