How can I train an "utterance" to display an answer to an intent in my Knowledge Graph?

I have 12 FAQs with configured questions and answers. I want to be able to have the bot understand what people might be asking based on one word in the question they are asking. Example:

The full question is: “Can I use Puffery (make sure the language used doesn’t require support) in all digital content?” If someone just types “Puffery”, somewhere in their question, I want the bot to say, "Did you mean? Can I use Puffery (make sure the language used doesn’t require support) in all digital content? "

Please advise. Thanks!

The “Did you mean” prompt will only appear for FAQs if the engine is either not confident over the response or there are multiple possibilities. If it appears that there is only a single clear question then the answer will be shown.

Now, you can engineer ambiguity through having multiple questions that have similar training, and perhaps that is what you want to achieve here.

An FAQ item has a primary question, a set of alternate questions and an answer. The primary question is just the first one in the list, but is significant in that this is the only one that is ever shown to the user. Therefore we generally recommend that this question to be correctly spelled and have good grammar. But the alternates are not shown, they are only used in the training and identification so they can contain all kinds of bad and partial phrasing.

In theory the questions and alternates are meant to represent what a user will say, and part of the matching process is to cross reference the user’s input to the question/alternate to see how similar they are. This means that the questions should be phrased in ways the user speaks, not as if the bot is asking the user a question or the user is reading a web page and interpreting a question.

No user speaking to the bot is ever going to say “make sure the language used doesn’t require support”. But by including it in the question then the platform will take all those words into account and conclude that “Puffery” is not very similar. A big clue if you are migrating on-line FAQs to conversation is if the question includes brackets or is series of “or” phrases.

Now you can get away with this in the primary question as a “sacrifice” because it can be shown to the user and that additional phrasing may help them. But you will probably need alternate training to account for what the user really says in different situations - that might be shorter, one word, utterances, or longer and winding utterances.

Also it is advisable for keywords to be called out in the KG as terms/tags, so that the platform knows that they are important and something to focus on.

Andy, thanks for this explanation…very clear!