Like nearly all of the Watson-branded offerings, the Watson Assistant is an application programming interface (API) hosted in the IBM Cloud. You learned how to create an account in the IBM Cloud in Chapter 1, Background, Transition, and the Future of Computing. We'll pick up from there by guiding you through the creation of an instance of the Watson Assistant service—referred to as a workspace—in which you will customize and train the service to your conversational context.

To get started, we need to create a workspace that Watson can work with:

1. After creating an account, go to the IBM Cloud services...

The first thing we need to do is identify the set of intents we want to support in this conversational agent. An intent is just an action that the end user may express, and that we want the application to support. Identifying the set of intents is actually pretty straightforward and can be derived from virtually any existing banking web application. 

This is a pretty common example of a typical set of actions you can perform on a banking application:

Here is the list of banking services that you might find on the Web:

• ACCOUNTS: Information on what accounts you have, and their balances, along with other information about the account
• MESSAGES: Where you would receive any notifications from the bank
• FUNDS TRANSFER: Enables you to move money between your own accounts
• MEMBER TO MEMBER: Supports transferring money from your account to...

Let's return to the first style of identifying an entity that you care about. In the first approach, we showed you in Recognizing entities, we went to the intent examples, and highlighted the entity that we cared about in that example. We refer to this as the annotator approach to entity recognition. Annotators are a much more powerful way of identifying the entities. Rather than setting a rule, such as the word hot is a temperature entity, the annotator approach builds a machine learning algorithm that takes in the context of the surrounding words in the sentence. In doing so, it calculates the probability that hot is referring to a temperature. In a sentence such as The water is hot, it makes sense to recognize hot as being a temperature. But in another sentence, such as The market is hot, it should be taken to be the relative...

To make your application a little more interesting, you can preface the interaction with a little greeting, and some sort of acknowledgement. We'll do that now.

You can introduce a dialog by pressing the Dialog tab of the main workspace screen and press the Create button. This will begin with a very basic dialog flow already filled out for you—including what to say to get things going, and how to respond in the case the user utters something completely off-topic:

Press on the Welcome card and an edit screen will come up on the right. You can modify the text that Watson will respond with, for example, change it to Welcome to ABC Bank. What can I do for you?:

Now, let's provide an interesting response if the user wants to pay their power bill. Press the Add node button, and in the If bot recognizes: field enter #Pay-Bills, press the + sign to...

So, now we're ready to start developing an application to make use of your trained Watson conversation. For this, we're going to use Python, and we're less concerned with the proper structuring of the application as a whole than with the structuring of your calls to the Watson Assistant services. However, before we begin, we need to get the Python SDK for Watson. If you don't already have Python installed, go to https://www.python.org/.

Once you have Python installed, you should be able to use the PIP tool to download and install the Watson SDK by issuing the following command in a terminal window:

pip install --upgrade ibm-watson

There are occasions where you may run into permission errors when you issue that command on Mac or Linux. If you do, type the following command:

sudo -H pip install --ignore-installed...

As we write this, IBM is developing some enhancements to the Watson Assistant service that will be available by the time this book is published. These include introducing the formal concept of an assistant. In essence, an assistant represents the persona facing your user set. Often an enterprise will give their virtual assistant a name – it's essentially an extension of their company brand. The assistant concept within the service serves as an aggregation point for the things that enable the brand's identity.

Generally, an assistant has a Skill—a formal reference to the set of Intents, Entities, and Dialogs that you created in your workspace. Any assistant you create is said to have that skill. But this formalism also makes deploying your assistant easier too. For example, having created an assistant, given it a name, and associated...

In this chapter, you learned about the Watson Assistant service. A good conversational agent is more than a single-turn, transactional response system. A conversation is about getting to the heart of the user's problem—keeping in mind that the first question someone asks is often not the question they really want to ask. Sometimes they don't even know what question to ask to solve their problem—they need to be walked through it. We've taught you the fundamentals of how to create a workspace in which you can train Watson to recognize the intents that your users might express, and to extract the entities within that expression. We also looked at how to build a dialog flow, in which you can guide the user through their conversation. We showed you how to jump around within that flow to address different ways that a user might proceed with their...