Introduction:
When your chatbot has been live and operational for a while, it will have had a lot of conversations. Some of them succesful, some of them not so. One of the things that will certainly happen, is that your user's question triggers an incorrect answer or flow in your chatbot. When that happens, that usually means that the utterance was matched to an incorrect intent.
So what can you do?
There are several things that you can do to optimise question recognition and letting your bot give the correct answers:
- Train your intents by adding extra sample utterances.
- Map user utterances to the right intent.
- Add missing intents and use cases to your bot. Because this is more a matter of design and organisation, we will not discuss that topic in this manual.
Map utterances to the right intent:
When you analyse the chatbot sessions in Analyze, you'll sometimes find that user questions are mapped to an incorrect intent, or to no intent at all. In this case, you can manually correct the mapping so that the utterance is associated with the right intent.
To map an utterance to another intent:
- In you chatbot session, navigate to the user utterance that you want to correct. In this example, you can see that the user didn't want to order a pizza, but wanted to know about the status of their pizza order.
- To correct this, go to the user utterance 'What is the status of my order', and click on the system.noresponse intent. This opens the Add utterances window.
- In the Add utterance window, select the intent to which you want to add the utterance. In our case the utterance 'What is the status of my order?' fits best with the intent system.status, which contains a flow where the user can leave their e-mail for up-to-date status information.
- Click Add to map the utterance to the selected intent.
Regularly analysing your chatbot sessions and correcting mismatched utterances, leads to significant accuracy of your chatbot!
Add extra utterances to an intent:
Your chatbot works with intents and natural language processing: an algorithm that analyses user utterances and links them to the correct bot flows and answers. It learns how to do this question - answer matching by learning from real-life utterances.
As you build your bot and add more and more intents, it's important that each intent has enough utterances to train it for that specific question. If this isn't the case, you bot might be less accurate in matching questions to the correct intent and therefore to the correct answer.
Adding extra utterances to an intent:
You can add utterances to your intent directly from within your dialog:
- Click the intent
- In the turn editor, add extra utterances.
- Save your intent.
When you want to batch edit a lot of intents, it's easier to add your utterances in Understanding > Intents.