Source: Deep Learning on Medium
Chatbot Training Automation and Acceleration with Variant Generator
Most of the chatbot platforms require defining intents and adding training phrases or utterances for every intent to train your chatbot. Dialogflow and LUIS recommend adding at least 10–15 utterances per intent. A chatbot cannot understand users accurately and identify correct intent if you add just a few utterances in intents. It is important to add as many training phrases as possible with wide variations to improve the performance of the chatbot in the real world.
Consider an example of a restaurant bot using which users can find and reserve a table in a restaurant. Users might say “show restaurants near me” or “view nearby restaurants” to view the available restaurants. To train such a bot effectively we have to add several variations that users might use to view the restaurants. How quickly can you come up with 15 unique variations? Following are three more variations that might come to your mind first:
- show restaurants around me
- restaurants in my neighborhood
- restaurants next to me
What if we have a tool that can auto-generate such variations? Fortunately, now we have one. You can try it for free using the link below.
It can generate several variations on the click of a button within a few seconds. All the variations are relevant and grammatically correct except a few which can be manually or automatically filtered using some similarity metrics but this tool can certainly help to automate and accelerate chatbot training. We can also quickly create a prototype of a chatbot using auto-generated utterances and then refine the training data in subsequent iterations.