You can create, tagg and manage your training sentences in the « Sentences » section.
This section gathers in one single place all the training sentences available in your project. These sentences are organized in a three-columns table: sentence, intent, labels.
The « sentence » column contains training sentences that can be tagged with entities.
The « intent » column displays the intent corresponding to the sentence. A sentence can have one and only intent at the same time. If the sentence has not yet been categorized, the text « Null » will be displayed, click on it to change its value.
The « labels » column displays all the labels attached to a specific sentence. A sentence can have zero, one or multiple labels at the same time. Labels enable you to define precisely how you want to organize your training data. Use it to add attributes to your sentences such as an origin (production, prototype, …) , a use (training, test, …), a characteristic (important, complex, no entity, …), …
By default, your sentences are organized according to their creation date. Sentences that have not been tagged with entities or an intent will appear first.
You can apply filters to the default view by clicking on the « Filters » button.
Filters can be used with the following elements:
- Intent confidence (only applied to sentences imported from production – see Import rules for more information)
- Nb of labels applied to a sentence
- Text in a sentence
The « entity » filter can be enriched with the following characteristics:
- Confidence (only applied to sentences imported from production – see Import rules for more information)
To enrich your « entity » filter, select entity, choose the number of entity you want to apply this filter on and click on the button.
You can then apply local criteria of type, value or confidence.
For example, if I want to display all sentences with the organize meeting intent and containing at least 2 entities with a type « WITH » and a value « Mickael » and « Sandy », I would need to apply the following combination of filters :
Multiple filters can be combined together to create a unique view of your data.
Once your filters applied, click on the « Create view » button to save this particular filter combination.
This view will then be easily accessible by clicking on the « Select view button »
Adding new sentences
Click on the « Add » button to add new sentences to your project.
After having written your sentences in the dedicated field, multiple options are available for finalizing the import:
- Click directly on the « save » button to import your sentences as it
- Use the « intent » button to select an intent that will be applied to all your sentences during the import
- Use the « labels » button to select one or multiple labels to apply to all your sentences
- Select the « parsing » option to automatically parse the sentences with the latest version of your model. This option is not available if you have not yet trained a model
Once your import options selected, click on save to add your sentences to your project. They will appear at the top of the table.
If you have chosen the « parsing » option you will need to validate the results of the parsing by clicking on the « Validate » button. This feature enables you to double-check the quality of the parsing before this sentences are available for the training.
Add an entity
To tag your sentences, simply start by clicking on the specific word or expression you want to tag. A tagging widget will appear and give you the ability to chose which entity and which value you want to tag the word with.
A word can be tagged with an entity OR with an entity and an associated value. It cannot be tagged with a value alone.
Create a new value
If the word or expression you want to tag doesn’t yet exist in your entity value table, you can create a new one by clicking on the button. The word will then be automatically added to the entity list of value.
Create a new synonym
If the word or expression you want to tag is a synonym of an existing value, select the corresponding value and click on the button to add this word to your list of synonyms. Click on validate to finalize the tagging of the word.