Linker

warning

This AI Task is only available if you added as a data source an RDF Graph!

To access this AI task click on AI Tasks and AI Assistant Linker:

You get to the following interface:

Create a Linker

To create a new Linker task click on Linkers .

You have to :

• name it
• add labels that will help to train it.

Labels are words or expressions that you expect to find in your data source that will help the linker to identify entities.

For example here we create a linker for the cocktail "Margarita" and we add labels like "Tequila", "Lime juice", "Triple sec", etc.

warning

This AI Task don't work with multilingual AI assistant. Make sure to choose the assistant's language when you create it.

Go to the Create AI Assistant section for more information on how to create an AI assistant.

Aggregate & Evaluate

Linker

Now that you have created a linker, you can choose to :

• Aggregate data to it
• Evaluate the data linked to it

Aggregate

On the Aggregate tab you can choose which instance of data to link to the linker you created.

You can include data by clicking on the corresponding line. It will alternatly pass from :

• true = exclude (means the data that matches the label you choose will be considered non relevant for the linker and these label will automatically be excluded of the train feature)
• false = excluded (means the data will be considered relevant for the linker)
• neutral = nothing (means the data will be considered neither relevant nor non relevant for the linker you will have to manually choose if you want to include it or not when training the linker)

Evaluate

• On the Evaluate tab you can see the labels you have assigned to the linker "Margarita" just created.

• You can click on each label to exclude or include them in the evaluation of the linker.

• If some labels added on creation are detected as not related to the data sources (rdf file) there will be a 0 in the match count column. For example here the label "Rhum" has no match in the cocktails dataset for "Margarita" linker.

• You necessarily have to exclude at least one label to be able to train the linker.

Finally you can click on Train & Save to see the evaluation results.

Train and Save

• Now you can see the training results of the linker you created. The score represents the quality of the linker based on the labels you provided.

• Here you can see that the linker has a 0.09 score meaning that it is not very good. It is because of the fact that it was excluded from the aggregate step on the line "Non-alcoholic Beverages".

Objective

Training a linker aims to use an API to reconcile labels with the linked entities.

You can find more information about the Linker API in the Linker API section.