As prerequisites to create NL using Kata-CLI, install Kata-CLI on your device and login with your account.
To create a new NL using kata-cli, first create a new folder for each NLU. After that, run the following command.
kata nl-init <nl_name>
It would create a new file
nlu.yml in which the nlu structure can be defined
# nl name name: nl_name # language (id | en) lang: id # visibility of your nl: public | private (default) visibility: public # Entities definition entities: # define the entity for this nlu someEntity: .... # define other entity for this nlu otherEntity: ...
Use push command to create and update the NLU
The following is an example on how to define entities through
nlu.yml file from Kata-CLI.
name: nl lang: id entities: # example for entity type 'dict' # need to define your own dictionary product: type: dict profile: default dictionary: baju: - baju - pakaian celana: - celana - cln - jeans # example for entity type 'phrase' ner: type: phrase profile: ner labels: - person - location # example using feature belongsTo quantity: type: phrase profile: number belongsTo: product # example for entity type 'trait' intent: type: trait profile: intent labels: - greetings - order - confirm - cancel # example root to other entity in same nl greetingsType: type: trait profile: default root: intent:greetings # <entity_name>:<label> labels: - hello - bye # example root to other entity in different nl faq: type: trait profile: faq root: kata:qisg/qisg:question # <nluId>/<entity_name>:<label> labels: - faq1 - faq2 - faq3 # example of using inherit sentiment: inherit: kata:sentiment/sentiment
Currently bulk training can only be done through CLI with the condition of max 100 sentences (100 lines) per file. Create a
.txt file and use following syntax to tag the sentence.
Saya mau pesan pizza #intent:order
Means the sentence is classified into entity intent as label order
Totalnya berapa ya? #intent:ask #questionType:how_much
Means the sentence is classified into entity intent as label order and entity questionType as label how_much
Saya mau pesan tiket ke (@destination Malang) atas nama (@ner:person Budi) #intent:order
Means the word “Malang” is tagged for entity destination and “Budi” is tagged for entity ner as label person
Pizza (@item-1 bbq) jumlahnya (@item-1.qty dua) ya #intent:order
Means the word “bbq” is tagged for entity item with id 1, “dua” is tagged for entity qty which belongs to entity item with id 1. Using id is required for defining belongsTo.
To train the model use the following training command. Give an interval for around 15-30 seconds for the model to completely finish training.
kata nl-train -f <file_name>
To test your NL, simply use the prediction command.
kata nl-predict -s <testing_sentence>