azure-ai-language-conversations

Azure Conversational Language Understanding client library for Python

Conversational Language Understanding - aka CLU for short - is a cloud-based conversational AI service which provides many language understanding capabilities like:

• Conversation App: It's used in extracting intents and entities in conversations
• Workflow app: Acts like an orchestrator to select the best candidate to analyze conversations to get best response from apps like Qna, Luis, and Conversation App
• Conversational Summarization: Used to analyze conversations in the form of issues/resolution, chapter title, and narrative summarizations

Source code | Package (PyPI) | Package (Conda) | API reference documentation | Samples | Product documentation | REST API documentation

Getting started

Prerequisites

• Python 3.7 or later is required to use this package.
• An Azure subscription
• A Language service resource

Install the package

Install the Azure Conversations client library for Python with pip:

pip install azure-ai-language-conversations

Note: This version of the client library defaults to the 2023-04-01 version of the service

Authenticate the client

In order to interact with the CLU service, you'll need to create an instance of the ConversationAnalysisClient class, or ConversationAuthoringClient class. You will need an endpoint, and an API key to instantiate a client object. For more information regarding authenticating with Cognitive Services, see Authenticate requests to Azure Cognitive Services.

Get an API key

You can get the endpoint and an API key from the Cognitive Services resource in the Azure Portal.

Alternatively, use the Azure CLI command shown below to get the API key from the Cognitive Service resource.

az cognitiveservices account keys list --resource-group <resource-group-name> --name <resource-name>

Create ConversationAnalysisClient

Once you've determined your endpoint and API key you can instantiate a ConversationAnalysisClient:

from azure.core.credentials import AzureKeyCredential
from azure.ai.language.conversations import ConversationAnalysisClient

endpoint = "https://<my-custom-subdomain>.cognitiveservices.azure.com/"
credential = AzureKeyCredential("<api-key>")
client = ConversationAnalysisClient(endpoint, credential)

Create ConversationAuthoringClient

Once you've determined your endpoint and API key you can instantiate a ConversationAuthoringClient:

from azure.core.credentials import AzureKeyCredential
from azure.ai.language.conversations.authoring import ConversationAuthoringClient

endpoint = "https://<my-custom-subdomain>.cognitiveservices.azure.com/"
credential = AzureKeyCredential("<api-key>")
client = ConversationAuthoringClient(endpoint, credential)

Create a client with an Azure Active Directory Credential

To use an Azure Active Directory (AAD) token credential, provide an instance of the desired credential type obtained from the azure-identity library. Note that regional endpoints do not support AAD authentication. Create a custom subdomain name for your resource in order to use this type of authentication.

Authentication with AAD requires some initial setup:

• Install azure-identity
• Register a new AAD application
• Grant access to the Language service by assigning the "Cognitive Services Language Reader" role to your service principal.

After setup, you can choose which type of credential from azure.identity to use. As an example, DefaultAzureCredential can be used to authenticate the client:

Set the values of the client ID, tenant ID, and client secret of the AAD application as environment variables: AZURE_CLIENT_ID, AZURE_TENANT_ID, AZURE_CLIENT_SECRET

Use the returned token credential to authenticate the client:

from azure.ai.language.conversations import ConversationAnalysisClient
from azure.identity import DefaultAzureCredential

credential = DefaultAzureCredential()
client = ConversationAnalysisClient(endpoint="https://<my-custom-subdomain>.cognitiveservices.azure.com/", credential=credential)

Key concepts

ConversationAnalysisClient

The ConversationAnalysisClient is the primary interface for making predictions using your deployed Conversations models. For asynchronous operations, an async ConversationAnalysisClient is in the azure.ai.language.conversation.aio namespace.

ConversationAuthoringClient

You can use the ConversationAuthoringClient to interface with the Azure Language Portal to carry out authoring operations on your language resource/project. For example, you can use it to create a project, populate with training data, train, test, and deploy. For asynchronous operations, an async ConversationAuthoringClient is in the azure.ai.language.conversation.authoring.aio namespace.

Examples

The azure-ai-language-conversation client library provides both synchronous and asynchronous APIs.

The following examples show common scenarios using the client created above.

Analyze Text with a Conversation App

If you would like to extract custom intents and entities from a user utterance, you can call the client.analyze_conversation() method with your conversation's project name as follows:

# import libraries
import os
from azure.core.credentials import AzureKeyCredential
from azure.ai.language.conversations import ConversationAnalysisClient

# get secrets
clu_endpoint = os.environ["AZURE_CONVERSATIONS_ENDPOINT"]
clu_key = os.environ["AZURE_CONVERSATIONS_KEY"]
project_name = os.environ["AZURE_CONVERSATIONS_PROJECT_NAME"]
deployment_name = os.environ["AZURE_CONVERSATIONS_DEPLOYMENT_NAME"]

# analyze quey
client = ConversationAnalysisClient(clu_endpoint, AzureKeyCredential(clu_key))
with client:
    query = "Send an email to Carol about the tomorrow's demo"
    result = client.analyze_conversation(
        task={
            "kind": "Conversation",
            "analysisInput": {
                "conversationItem": {
                    "participantId": "1",
                    "id": "1",
                    "modality": "text",
                    "language": "en",
                    "text": query
                },
                "isLoggingEnabled": False
            },
            "parameters": {
                "projectName": project_name,
                "deploymentName": deployment_name,
                "verbose": True
            }
        }
    )

# view result
print("query: {}".format(result["result"]["query"]))
print("project kind: {}\n".format(result["result"]["prediction"]["projectKind"]))

print("top intent: {}".format(result["result"]["prediction"]["topIntent"]))
print("category: {}".format(result["result"]["prediction"]["intents"][0]["category"]))
print("confidence score: {}\n".format(result["result"]["prediction"]["intents"][0]["confidenceScore"]))

print("entities:")
for entity in result["result"]["prediction"]["entities"]:
    print("\ncategory: {}".format(entity["category"]))
    print("text: {}".format(entity["text"]))
    print("confidence score: {}".format(entity["confidenceScore"]))
    if "resolutions" in entity:
        print("resolutions")
        for resolution in entity["resolutions"]:
            print("kind: {}".format(resolution["resolutionKind"]))
            print("value: {}".format(resolution["value"]))
    if "extraInformation" in entity:
        print("extra info")
        for data in entity["extraInformation"]:
            print("kind: {}".format(data["extraInformationKind"]))
            if data["extraInformationKind"] == "ListKey":
                print("key: {}".format(data["key"]))
            if data["extraInformationKind"] == "EntitySubtype":
                print("value: {}".format(data["value"]))

Analyze Text with an Orchestration App

If you would like to pass the user utterance to your orchestrator (worflow) app, you can call the client.analyze_conversation() method with your orchestration's project name. The orchestrator project simply orchestrates the submitted user utterance between your language apps (Luis, Conversation, and Question Answering) to get the best response according to the user intent. See the next example:

# import libraries
import os
from azure.core.credentials import AzureKeyCredential
from azure.ai.language.conversations import ConversationAnalysisClient

# get secrets
clu_endpoint = os.environ["AZURE_CONVERSATIONS_ENDPOINT"]
clu_key = os.environ["AZURE_CONVERSATIONS_KEY"]
project_name = os.environ["AZURE_CONVERSATIONS_WORKFLOW_PROJECT_NAME"]
deployment_name = os.environ["AZURE_CONVERSATIONS_WORKFLOW_DEPLOYMENT_NAME"]

# analyze query
client = ConversationAnalysisClient(clu_endpoint, AzureKeyCredential(clu_key))
with client:
    query = "Reserve a table for 2 at the Italian restaurant"
    result = client.analyze_conversation(
        task={
            "kind": "Conversation",
            "analysisInput": {
                "conversationItem": {
                    "participantId": "1",
                    "id": "1",
                    "modality": "text",
                    "language": "en",
                    "text": query
                },
                "isLoggingEnabled": False
            },
            "parameters": {
                "projectName": project_name,
                "deploymentName": deployment_name,
                "verbose": True
            }
        }
    )

# view result
print("query: {}".format(result["result"]["query"]))
print("project kind: {}\n".format(result["result"]["prediction"]["projectKind"]))

# top intent
top_intent = result["result"]["prediction"]["topIntent"]
print("top intent: {}".format(top_intent))
top_intent_object = result["result"]["prediction"]["intents"][top_intent]
print("confidence score: {}".format(top_intent_object["confidenceScore"]))
print("project kind: {}".format(top_intent_object["targetProjectKind"]))

if top_intent_object["targetProjectKind"] == "Luis":
    print("\nluis response:")
    luis_response = top_intent_object["result"]["prediction"]
    print("top intent: {}".format(luis_response["topIntent"]))
    print("\nentities:")
    for entity in luis_response["entities"]:
        print("\n{}".format(entity))

Conversational Summarization

You can use this sample if you need to summarize a conversation in the form of an issue, and final resolution. For example, a dialog from tech support:

# import libraries
import os
from azure.core.credentials import AzureKeyCredential
from azure.ai.language.conversations import ConversationAnalysisClient
# get secrets
endpoint = os.environ["AZURE_CONVERSATIONS_ENDPOINT"]
key = os.environ["AZURE_CONVERSATIONS_KEY"]
# analyze query
client = ConversationAnalysisClient(endpoint, AzureKeyCredential(key))
with client:
    poller = client.begin_conversation_analysis(
        task={
            "displayName": "Analyze conversations from xxx",
            "analysisInput": {
                "conversations": [
                    {
                        "conversationItems": [
                            {
                                "text": "Hello, how can I help you?",
                                "modality": "text",
                                "id": "1",
                                "participantId": "Agent"
                            },
                            {
                                "text": "How to upgrade Office? I am getting error messages the whole day.",
                                "modality": "text",
                                "id": "2",
                                "participantId": "Customer"
                            },
                            {
                                "text": "Press the upgrade button please. Then sign in and follow the instructions.",
                                "modality": "text",
                                "id": "3",
                                "participantId": "Agent"
                            }
                        ],
                        "modality": "text",
                        "id": "conversation1",
                        "language": "en"
                    },
                ]
            },
            "tasks": [
                {
                    "taskName": "Issue task",
                    "kind": "ConversationalSummarizationTask",
                    "parameters": {
                        "summaryAspects": ["issue"]
                    }
                },
                {
                    "taskName": "Resolution task",
                    "kind": "ConversationalSummarizationTask",
                    "parameters": {
                        "summaryAspects": ["resolution"]
                    }
                },
            ]
        }
    )

    # view result
    result = poller.result()
    task_results = result["tasks"]["items"]
    for task in task_results:
        print(f"\n{task['taskName']} status: {task['status']}")
        task_result = task["results"]
        if task_result["errors"]:
            print("... errors occurred ...")
            for error in task_result["errors"]:
                print(error)
        else:
            conversation_result = task_result["conversations"][0]
            if conversation_result["warnings"]:
                print("... view warnings ...")
                for warning in conversation_result["warnings"]:
                    print(warning)
            else:
                summaries = conversation_result["summaries"]
                print("... view task result ...")
                for summary in summaries:
                    print(f"{summary['aspect']}: {summary['text']}")

Import a Conversation Project

This sample shows a common scenario for the authoring part of the SDK

import os
from azure.core.credentials import AzureKeyCredential
from azure.ai.language.conversations.authoring import ConversationAuthoringClient

clu_endpoint = os.environ["AZURE_CONVERSATIONS_ENDPOINT"]
clu_key = os.environ["AZURE_CONVERSATIONS_KEY"]

project_name = "test_project"

exported_project_assets = {
    "projectKind": "Conversation",
    "intents": [{"category": "Read"}, {"category": "Delete"}],
    "entities": [{"category": "Sender"}],
    "utterances": [
        {
            "text": "Open Blake's email",
            "dataset": "Train",
            "intent": "Read",
            "entities": [{"category": "Sender", "offset": 5, "length": 5}],
        },
        {
            "text": "Delete last email",
            "language": "en-gb",
            "dataset": "Test",
            "intent": "Delete",
            "entities": [],
        },
    ],
}

client = ConversationAuthoringClient(
    clu_endpoint, AzureKeyCredential(clu_key)
)
poller = client.begin_import_project(
    project_name=project_name,
    project={
        "assets": exported_project_assets,
        "metadata": {
            "projectKind": "Conversation",
            "settings": {"confidenceThreshold": 0.7},
            "projectName": "EmailApp",
            "multilingual": True,
            "description": "Trying out CLU",
            "language": "en-us",
        },
        "projectFileVersion": "2022-05-01",
    },
)
response = poller.result()
print(response)

Optional Configuration

Optional keyword arguments can be passed in at the client and per-operation level. The azure-core reference documentation describes available configurations for retries, logging, transport protocols, and more.

Troubleshooting

General

The Conversations client will raise exceptions defined in Azure Core.

Logging

This library uses the standard logging library for logging. Basic information about HTTP sessions (URLs, headers, etc.) is logged at INFO level.

Detailed DEBUG level logging, including request/response bodies and unredacted headers, can be enabled on a client with the logging_enable argument.

See full SDK logging documentation with examples here.

import sys
import logging
from azure.core.credentials import AzureKeyCredential
from azure.ai.language.conversations import ConversationAnalysisClient

# Create a logger for the 'azure' SDK
logger = logging.getLogger('azure')
logger.setLevel(logging.DEBUG)

# Configure a console output
handler = logging.StreamHandler(stream=sys.stdout)
logger.addHandler(handler)

endpoint = "https://<my-custom-subdomain>.cognitiveservices.azure.com/"
credential = AzureKeyCredential("<my-api-key>")

# This client will log detailed information about its HTTP sessions, at DEBUG level
client = ConversationAnalysisClient(endpoint, credential, logging_enable=True)
result = client.analyze_conversation(...)

Similarly, logging_enable can enable detailed logging for a single operation, even when it isn't enabled for the client:

result = client.analyze_conversation(..., logging_enable=True)

Next steps

More sample code

See the Sample README for several code snippets illustrating common patterns used in the CLU Python API.

Contributing

See the CONTRIBUTING.md for details on building, testing, and contributing to this library.

This project welcomes contributions and suggestions. Most contributions require you to agree to a Contributor License Agreement (CLA) declaring that you have the right to, and actually do, grant us the rights to use your contribution. For details, visit cla.microsoft.com.

When you submit a pull request, a CLA-bot will automatically determine whether you need to provide a CLA and decorate the PR appropriately (e.g., label, comment). Simply follow the instructions provided by the bot. You will only need to do this once across all repos using our CLA.

This project has adopted the Microsoft Open Source Code of Conduct. For more information see the Code of Conduct FAQ or contact [e-mail] with any additional questions or comments.

Release History

1.1.0 (2023-06-13)

Features Added

• Added support for service version 2023-04-01.

Breaking Changes

Note: The following changes are only breaking from the previous beta. They are not breaking since version 1.0.0 when those types and members did not exist.

• Removed support for service version 2022-05-15-preview.
• Removed support for service version 2022-10-01-preview.
• Removed support for "ConversationalPIITask" analysis with ConversationAnalysisClient.
• Removed support for "ConversationalSentimentTask" with ConversationAnalysisClient.
• Removed the following methods from ConversationAuthoringClient:
  • begin_assign_deployment_resources
  • get_assign_deployment_resources_status
  • begin_unassign_deployment_resources
  • get_unassign_deployment_resources_status
  • begin_delete_deployment_from_resources
  • get_deployment_delete_from_resources_status
  • list_assigned_resource_deployments
  • list_deployment_resources

1.1.0b3 (2022-11-10)

Features Added

• Added support for the "ConversationalSentimentTask" kind with begin_conversation_analysis.
• Added support for "chapterTitle" and "narrative" summaryAspects options for ConversationalSummarizationTasks.
• Added methods to the ConversationAuthoringClient to manage deployment resources:
  • begin_assign_deployment_resources
  • get_assign_deployment_resources_status
  • begin_unassign_deployment_resources
  • get_unassign_deployment_resources_status
  • begin_delete_deployment_from_resources
  • get_deployment_delete_from_resources_status
  • begin_load_snapshot
  • get_load_snapshot_status
  • list_assigned_resource_deployments
  • list_deployment_resources
• Added optional trained_model_label keyword argument to begin_export_project.

Other Changes

• This version and all future versions will require Python 3.7+. Python 3.6 is no longer supported.

1.1.0b2 (2022-07-01)

Features Added

• Added Azure Active Directory (AAD) authentication support
• Added support for authoring operations with ConversationAuthoringClient under the azure.ai.language.conversations.authoring namespace.

1.0.0 (2022-06-27)

Features Added

• Added Azure Active Directory (AAD) authentication support
• Added more resolution types for entities
• Added support for authoring operations with ConversationAuthoringClient under the azure.ai.language.conversations.authoring namespace.

Breaking Changes

• Client now uses python dictionaries for method parameters and results instead of classes.

1.1.0b1 (2022-05-26)

Features Added

• Conversation summarization task (Long-running operation)
• Conversation PII extraction task (Long-running operation)

Breaking Changes

• Client now uses python dictionaries for method parameters and results instead of classes.
• Many input and result parameter name changes in analyze_conversation() method

1.0.0b3 (2022-04-19)

Features Added

• Entity resolutions
• Extra features

Breaking Changes

• The ConversationAnalysisOptions model used as input to the analyze_conversation operation is now wrapped in a CustomConversationalTask which combines the analysis options with the project parameters into a single model.
• The query within the ConversationAnalysisOptions is now further qualified as a TextConversationItem with additional properties.
• The output AnalyzeConversationResult is now wrapped in a CustomConversationalTaskResult according to the input model.

Other Changes

• Python 2.7 is no longer supported. Please use Python version 3.6 or later.

1.0.0b1 (2021-11-03)

Features Added

• Initial release