Machine Learning Made Easy: How to Use Azure Cognitive Services to Train Bots


April 20, 2022

With API-led automation and Robotic Process Automation (RPA) making automated solutions possible for citizen developers, now is the ideal time to start getting comfortable with the low-code and no-code aspects of this field. 

Of course, most automation solutions are more complex than some providers will have you believe. Like the bots themselves, humans need training and guidance to be able to use this form of automation successfully. Even experienced IT professionals require training to be able to work with no-code and low-code platforms effectively, so don’t worry if it takes you a bit of time to get comfortable. 

Perhaps the best place to start your first adventure into the world of automation is with the service bots that interact with your customers. Through AI training, you can remove the frustrating aspects of interacting with a bot while boosting the level of service they’re able to provide to your customers. As you can imagine, highly efficient bots that are sensitive to slight variations in user intent are a valuable asset for any business.

AI Training with Azure Cognitive Services

Answering questions is one of the core functions a customer service bot must serve. Whether your bot covers just a few basics or a complex range of possible enquiries, it must have a question and answer framework guiding it. This is where Microsoft Azure Cognitive Services comes in – specifically, the QnA Maker.

Microsoft Cognitive Services QnA Maker

This Natural Language Processing (NLP) tool is free, cloud-based, and surprisingly easy to use (once you get the hang of it). It can help you train your bots to respond to customer queries in a way that feels natural, conversational, and not the least bit infuriating. This makes it ideal for training chatbots, social media apps, and speech-enabled desktop apps. 

The simplicity of this tool lies in the fact that you may already have all the material it needs to create its Knowledge Base (also known as KB, but let’s not go overboard with the acronyms). By extracting data from the FAQs on your website or a document you provide, QnA Maker can automatically extract, train, and publish an effective Knowledge Base in the time it takes you to make a cup of coffee.

Training AI with QnA Maker

The QnA Maker works its magic via the three steps outlined above: extraction, training, and publishing.

1. Extracting content with QnA Maker

The first step in creating your Knowledge Base is to extract the relevant content. When you create a new service in QnA Maker, all you have to do is direct it to Add sources for your KB and then click on create. The tool can also automatically extract data from your documents and your FAQ URLs. You’ll also have the power to edit, remove, and add data manually, allowing you to go deep in fine-tuning your Knowledge Base. 

From here, it’s on to the fun part: testing and training the AI.

2. Training AI with QnA Maker

While it’s essential to start with a robust and thorough Knowledge Base, in many ways the training of the AI is the most important step in this process. This is where you get to evaluate how correct your responses are, and crucially, where you have the opportunity to retrain the bot if it isn’t hitting the mark. 

The key factor you’re looking to achieve is relevance. Let’s take a travel agency bot for example. If you ask, “How do I book a hotel transfer?”, and it directs you to the answer for “How do I book a hotel?”, then you know you need to add to your Knowledge Base and retrain the bot to pull the correct answer when given this configuration of words. Note that a very slight change in words (in the case above, just one additional word at the end of the question) can drastically change the user’s intent. So, it’s essential to take your time during this process to ensure you create a bot that can accurately address queries and send your customers in the right direction. 

Hint: Always press save and retrain after making any changes or adding any content to your Knowledge Base.

If you have a bot that’s already live, another helpful feature of QnA Maker is the ability to download its chat logs and use this as a guide for reworking your Knowledge Base and retraining your bot. When you hit Download Chat Logs, you’ll get a list of all the questions being submitted to the API end-point. The log automatically orders the questions by frequency, making it easier to see what customers want and whether their needs are being met. From here, you can determine which questions need fine-tuning and retraining. 

3. Publishing your bot

The Azure QnA Maker gives you the opportunity to preview the adjustments you’ve made and the training you’ve done to see how it impacts the activity of the bot. To do this, you simply need to download the diff file. Once you’re happy, it’s simply a matter of clicking Publish

For more information on Azure Cognitive Services or how Melbourne businesses can benefit from QnA Maker, feel free to reach out to Invotec. Call 1300 468 683 or fill out the form below, and you’ll be connected to an experienced IT consultant who’ll be happy to share their expertise with you. 

[cboxarea id=”cbox-zCMaBzoelTgg3AcK”]

Book a FREE Consultation

When you choose Invotec, we want you to feel 100% confident. That’s why we offer a free consultation for all schools, to see if we’re a perfect fit. Request your free consultation today and take the first step towards better IT Support.

This field is for validation purposes and should be left unchanged.