/model-api-example

Primary LanguageKotlin

Model API

​ Model API is a JAICP feature that allows our partners to use third-party tokenizers, named entity- and intent-recognition services together with JAICP DSL.

Contract

The third-party service implementing the Model API feature should fulfill a contract that is described in Model API Specification.

Integration with JAICP project

To use external NLU service in your JAICP project you should configure it in project NLU settings. Please, refer to the documentation.

Service example

This repository contains an example of a microservice implementing Model API specification using Spacy as tokenizing engine and Dialogflow as an intent-processor and named entity recognizer (NER). ​

Example restrictions:

Due to Dialogflow restrictions, given example only processes named entities present inside valid intents and does not support topN intents feature, providing single best-match intent (thus, is only capable of correctly handling global intents). ​

Requirements:

Following requirements will be installed automatically if you choose to run the service in docker container:

  • Java 8;
  • Maven;
  • Python3;
  • Spacy (And SudachiPy for Japanese language processing). ​

Installation

  • Create a dialogflow project;
  • Fill in the intents, each intent must have a single output context with a lifetime of 1 request;
  • Create a service account in a Google Cloud Platform project associated with Dialogflow project;
  • Download the service account key as JSON and place it inside the microservice credentials folder (by default it's named "creds")
  • Fill in the configuration files (description below)
  • Run in docker:
    • Run mvn package
    • Run docker compose up
  • Run without docker:
    • Run mvn spring-boot:run ​

Configuration

application.properties:
model.api.auth.login    //Basic auth credentials used to access the microservice
model.api.auth.password
dialogflow.properties:
dialogflow.nerProjectId         //GCP id of a project that will be used for NER
dialogflow.credentialsDirectory //Path to a folder containing service key JSONs
exceptions.properties:
exception.message.malformedBody     //Map of language codes to error messages for requests with malformed bodies

preprocessor.spacyPath          //Path to main spacy script
preprocessor.modelByLanguage    //Map of language codes to spacy model names