Converse AI by Smartsheet

Phrases are the individual words or sets of words that are submitted by you to each individual template, to help train it how to match it against individual user requests.

You can submit phrases in plain text, without needing to understand anything about NLP, and Converse will learn from them accordingly. More Advanced Phrase options are available if required though.

Every time you add a phrase to Converse, it parses it to understand what it can from the phrase, and to teach its database accordingly.

It then uses both it's database of Contexts and it's own learning algorithms to match multiple other inputs based on any given phrase, meaning that you can teach it to cope with a wide variety of inputs with as little as 5-10 phrases.

If you have specific terminology or wording related to your service, you can also add your own Contexts to enable Converse AI to improve it's matching capabilities.

As well as just using simple text to teach Converse AI, there are a number of more advanced options if required:

Optional Words

By specifying optional words within sets of {} Converse is able to match multiple versions automatically, e.g:

Can you find me {another} one?

Will match:
Can you find me another one? and Can you find me one

This or That

By specifying single words with a | between them (no spaces) Converse AI will match either one, e.g

Can you find me|us another one? will match:

Can you find me another one? and Can you find us another one.

Of course you can also use these together, so

Can you find me|us {another} one? will match all four possible combinations.

Contexts within Phrases

Although Converse AI will automatically learn and match Contexts within phrases, if you would like to manually specify them for tighter matching you can, simply by using them inline e.g:

I need @sys-colors paint

Pattern Matching

If you don't want to use natural language parsing, if you have simpler requirements, you can also use patterns.

These include

contain:something will match any phrase that has the word something in it.
regex:yourregexhere` will match a particular regular expression.

These are tested after any natural language matching occurs, so can be used for simplicity or fallback as required.


What about if you need to specify the same context in a single phrase twice, in that case you can use roles to ensure the information is extracted correctly, e.g:

I need to get from @sys-uscities:origin to @sys-uscities:destination

Updated 2 years ago


Suggested Edits are limited on API Reference Pages

You can only suggest edits to Markdown body content, but not to the API spec.