Our Training tool makes it easy and fast to create new models, that will help Smartbnb detect questions. This means you are able to train Smartbnb to detect any question that would go through your mind, such as questions about stairs, beach, towels, deposits...
Training to detect questions works hand in hand with the Questions flow:
- You create your model to detect a new category of questions.
- You can create your reply to answer to those categories of questions.
How does it work?
Smartbnb detects questions on the basis of AI models. Those models are a collection of examples relevant to the same domain. For example, our model about "parking" contains hundreds of examples of sentences, that should be classified as questions about parking.
Models are "training" data. Using those examples, provided by humans, the AI will be able to infer probabilities when a new, unplanned, question comes up.
This is when the "AI magic" happens. You don't need to provide all the many ways of asking exactly the same questions; you just need to provide sufficient examples for the AI to use the model to draw "its own" conclusions.
What is a custom model?
Custom models are models that you can create using Smartbnb's Training tool.
From there, you can:
- create your new model,
- populate it with real-life examples,
- test that the detection matches your expectations.
We will also let you know about the "health" of your model: the more comprehensive a model it is, the better it is at drawing conclusions for questions outside the model.
Once a model is strong enough, Smartbnb will be able to classify in this category messages that are not exact matches to the ones provided in the model but are similar enough.
By default, the detection is "shy", which means:
- a weak model will only detect near perfect matches with the examples provided,
- a stronger model will mean a very accurate detection outside the model.
How do custom models articulate with Smartbnb's managed models?
Smartbnb does manage its own models, that are available to every user, no setup required.
Our models are however more complex, and have a higher "priority" than custom models: if a question is a match of a Smartbnb managed model, then custom models will not be checked for the specific sentence and/or message.
What happens when several Smartbnb users work on the same questions?
We want to make it easy for everyone to automate everything. When several Smartbnb users are working on the same domain, the models are "pooled" together to provide an enhanced detection, beyond your own, personal model.
For example, several hosts may want to automate questions relevant to deposits. If any custom model about deposits are a match, it's considered a match for all models about deposits.
How to create a custom model?
You can create your custom model using Smartbnb's Training tool.