Support > Documentation > Examples > NLP Service Comparison

NLP Service Comparison

Overview

Demonstrates the integration of cloud-based Natural Language Processing services for comparison within a Q & A format.

Page Contents

ExampleNLP
ExampleNLP

Purpose

Shows how to leverage commodity NLP AI services through REST calls, using RIDE’s common web services system.

Leverages RIDE’s generic NLP C# interface, primarily focused on question-answering interactions, with the principle generally working for user input and agent output. Optionally includes sentiment analysis and entity analysis.

Current Providers

  • Microsoft Azure Cognitive Service for Language
  • Amazon (AWS) Lex
  • Google DialogFlow
  • OpenAI GPT-3
  • OpenAI ChatGPT-4

How to Use

IMPORTANT

If you wish to use the ExampleNLP scene and related web services, you must update your local RIDE config file (ride.json). Please back-up any local changes to the file, including your terrain key. Next, restore your Config to default using the corresponding option in the Config debug menu at run-time, then reapply any customization and terrain key. Refer to the Config File page within Support for more information. 

Test the different types of inputs, different lengths of responses, and measure metrics as the average response time for both “hot” and “cold” services.

A service is said to be “cold” when the service requesting the NLP agent makes the very 1st request just after its initialization and deployment and the agent has not been used for a while (here leaving all the respective platform agents idle for >=5 minutes).

When the service requesting the NLP agent has already made very 1st request just after its initialization and deployment, any subsequent service request is considered “hot”.

Single Provider UI

  1. Click Cycle Providers button to cycle through currently available services.
  2. Input text in the bottom field and hit enter or click the Ask button.
  3. The side pane populates with the following information after entering a query:
    1. Response Time
    2. Text Analytics
    3. Entities
  4. If present, a custom or automatic answer will appear between the pane and input field.

Note that most services are domain specific, with just a handful of questions and answers authored, purely for demonstration purposes. Example question: “What is RIDE?” The exception is OpenAI GPT-3, which is a general purpose language model.

Ask the same question and see how the output differs.

Multiple Provider Comparison UI

  1. Click the Compare Providers button to toggle the live, side-by-side comparison interface.
  2. Input text in the bottom field and hit enter or click the Ask button. Any question will be sent to all 3 providers simultaneously.
  3. Each service has its own pane which populates with the following information after entering a query:
    1. Answer
    2. Response Time
    3. Text Analytics
    4. Entities

Scene Location & Name

Assets/Ride/Examples/NLP/ExampleNLP.unity

Setup Requirements 

The ExampleNLP scene utilizes a customizable UI and interchangeable service providers through canvas, scripts and prefabs. Explore the objects in the Hierarchy view for the scene inside the Unity editor and source for Ride.NLP via the API documentation.

The main script is ExampleNLP.cs in the same folder.

Any NLP option implements the INLPQnASystem C# Interface, which itself is based on INLPSystem. These systems provide the main building blocks to interface with an NLP service:

				
					
OpenAIGPT3System m_openAIGPT3;
var openAIGPT3Component = new GameObject("OpenAIGPT3System").AddComponent<OpenAIGPT3System>();
m_openAIGPT3 = openAIGPT3Component;

The specific endpoint and authentication information is defined through RIDE’s configuration system:

				
					
m_configSystems = Globals.api.systemAccessSystem.GetSystem<RideConfigSystem>();
openAIGPT3Component.m_uri = m_configSystems.config.openAI.endpoint;
openAIGPT3Component.m_authorizationKey = m_configSystems.config.openAI.endpointKey;

The NLP system encapsulates the user question into an NLPQnAQuestion or NLPRequest object. This user input is sent through the RIDE common web services system. The results are encapsulated into an NLPQnAAnswer  or NLPResponse object. A provided callback function allows a client (in this case this ExampleNLP) what specifically to do with the result. The NLP system itself handles much of the work, so that the client only requires one line to send the user question: 

				
					
m_openAIGPT3.AskQuestion(userInput, OnCompleteAnswer);

The project/effort/work depicted here was or is sponsored by the U.S. Army Research Laboratory (ARL) under contract number W911NF-14-D-0005. Statements and opinions expressed and content included do not necessarily reflect the position or the policy of the Government, and no official endorsement should be inferred.