Data augmentation is a strategy that enables practitioners to significantly increase the diversity of data available for training models, without actually collecting new data.
In the illustration below, we can see how a single document can be slighly changed to generate many new documents that can be really usefull to feed a machine learning model.
We have seen above how we can generate new images programatically, guess what, the same is possible with text!
Using different techniques, typos can can be introduced in your intents, leading to more robustness to the said typos.
Our bot platform offers a unique data augmentation component that can improve the robustness of the machine learning models of your chatbots.
To use go to your bot settings and pick a model featuring
data augmentation save the settings and train your bot.
Currently, the only available model featuring data augmentation is
By default, our component will generate 3 variations per utterance.
We plan to offer you the ability to define this number.
Note that the only drawback of this component is the increase of the training time, otherwise it can only get better!
Updated 10 days ago