Page 15 - MSDN Magazine, December 15, 2017

Figure 6 Top Bots Being Used in Skype
monitor your models at scale. Whether deploying to the cloud, to a data lake, to a database like SQL Server 2017 or to the intelligent edge, Azure Machine Learning provides outstanding flexibility for model deployment.
Azure Machine Learning lets you use the toolkits you’re familiar with—Cognitive Toolkit (CNTK), Tensorflow, Caffe and more.
Once you’ve created the Azure Machine Learning Experimentation and Model Management Service, you can download the Machine Learning Workbench to help with your data preparation tasks, as well as manage all your machine learning projects.
Building Bots
Companies have started using intelligent bots to enable users to interact with their services. Bots provide tremendous value in a host of scenarios, including customer sup- port engagements, answering product-related questions, providing tourists with navigation instructions or respond- ing to HR-related questions. We see bots being used in exciting new ways and in different industries, including retail, health care, manufacturing, telecommunications and the public sector.
For example, the AzureBot lets you manage your Azure services (for instance, starting or stopping Azure VMs) using a bot that’s available on Skype, Microsoft Teams and
other platforms. The Summarize bot uses Bing to present the main points from any Web page. Figure 6 shows some of the top bots currently being used in Skype (bing.com/search?q=top+bots).
You can build an intelligent bot in minutes using the Azure Bot Service, then deploy it to reach customers on multiple channels, like Skype, Messenger, Microsoft Teams or via a Web site.
To jumpstart development of your first bot, leverage the QnA Maker, a service that makes it easy to create a question and answer bot from an existing frequently asked questions (FAQ) page. Under- neath the hood, QnA maker uses state-of-the-art machine learning algorithms and natural language processing (NLP) to distill the information found on FAQ pages to create question-and-answer pairs, as depicted in Figure 7. Visit qnamaker.ai to find out more.
Let’s look at the architecture for building an intelligent bot that’s infused with a range of AI capabilities. These include Cognitive
The main components of Azure Machine Learning are: • Azure Machine Learning Workbench: Consists of a desktop application and command-line tools that let you manage the entire data science lifecycle, from data ingestion and preparation to model development and deployment. Machine Learning Workbench is available on both Windows and macOS. Figure 5
shows the Machine Learning Workbench UI.
• Experimentation Service: Runs training models across different machine learning environments, ranging from a local machine to a Docker container on a remote virtual machine (VM) to a scaled-out Spark cluster on Azure in the cloud. Experimen- tation Service integrates with Machine Learning Workbench and supports Git integration, access
control and sharing among co-workers.
• Model Management Service: Enables manage- ment and deployment of machine learning workflows and models. The Model Management Service keeps track of different model versions, and can package and deploy machine learning models as REST APIs
served from a Docker container.
Tech Help Bot
Entertainment Bot
Health Bot
Recipe Bot
. . .
Chat Channels E-Mail FB Skype Slack Web . . . Bot Platforms Microsoft Bot Framework Other Bot Platforms
msdnmagazine.com
Dec. 15, 2017 / Connect(); Special Issue 11
Figure 7 QnA Maker Architecture
Knowledge Base Endpoint
Indexing & Ranking Extract QnA
url pdf
QnA Maker
FAQ Content
txt
doc