watson_chatbot

creating a watson chatbot using IBM watson services for a flower shop

Do's and Don'ts

In terms of a good chatbot design there are three fundamental rules that you should follow

First Rule

Avoid using YES or NO in your replies

If your chatbot fails to interpret the question, a yes or no answer would be misleading or provide wrong information.

For example:

We might set the reply for a question that asks for

Q. Is delivery free?
A. Yes! it is

But what happens when the question is changed slightly

Q. Is delivery free or do I have to pay for it?
A. Yes! it is

The same answer won't work here. This situation arrived because of the negatives included in the question, where a human would answer no our chatbhot ends up answering yes

It would be far better to provide an answer that covers the topic of delivery price, without including an element of yes or no in it.

Correct Reply

Q. Is delivery free or do I have to pay for it?
A. Deliveries are free.

Second Rule

Incorporate part of user's question in your response whenever possible

For example, say

A. "I understand that you’re looking for help with an order.
	Unfortunately, I'm unable to look into orders that have already been placed."

And then perhaps we would have the chatbot refer the customer to our phone number or email address, in order to help them perhaps speak with a live person about their concern. This will further cement, in our user's mind, the relative intelligence of our chatbot, even though it technically failed to directly provide what the customer was hoping to find out.

Third Rule

Succinct and accurate answers are the best

This means we want to be informative but not reply with a wall of text. So we should keep our answers short. When a wall of text is an absolute must, to convey the information,then deflecting by sending the user to an informative page is a better approach.

Creating Watson Chatbot

Register with Watson Assistant by clicking on this link.
Verify your account by clicking the confirmation email you receive by email (be sure to check your spam folder, if you don't see the email).
Log in to IBM Cloud (which hosts the Watson Assistant service) and when prompted, chose a cloud region and create an organization and space. US-South will work for many, but you have the option to select other cities and countries. For the organization, you can use your company name or even your own name. For space, pick a namespace that makes sense to you. For example, it can be a department in your organization (e.g., marketing) or a phase of development (test, dev, staging, or prod). You'll be able to customize and add new regions, organizations, and spaces at any time, so don't worry too much about what you select at first.
Click on the Assistant service within the Watson portion of the IBM Cloud catalog. This will bring you to a service creation page where you'll be asked to enter a name for the service. You can enter whatever you like or simply watson-assistant-flower-shop Select the Lite plan that offers you up to 10,000 requests per month for free and click on Create Click on the Launch tool to launch the Watson Assistant user interface.
Click on the workspace to create any workspace or click on the existing one, and explore the example chatbot. What intents and entities are there? If you are adventurous you can even take a look at the dialog section. Play around with the default example in Watson Assistant and when you feel like you have become more familiar with it, go back to the workspaces list by clicking on the table icon in the left side of the page.

Creating Workspace

Once we have our watson assistant ready, go to the workspace page again and click on Create Workspace You will be asked for a name, I named it Florence Chatbot then enter the description for my case I used

Flower Shop Chatbot for providing assistance in flower suggestions and delivery 
information.

Choose your preferred language for chatbot and Create.

The redirected page opened says to Create Intents because our chatbot currently has no intents defined. On the top right corner click on Try it out, this is where you can test your chatbot after each progress. You can enter in any message to chatbot.

Let's enter hello and we have no response and the bar under our message is also empty, the bar will display the intent of the message.

Create Intent

On the Intent page click on Add Intent and a form appears, name your intent greetings and add in description like greet the user and in the Add user examples add the examples of greetings.

For example

hello
hi
hey
hey there
what's up
how you doing
good morning
good evening
greetings
nice to see you
...

You can add as many intents you like based on many ways a user might greet our bot. Once added intents you might see that Try it out says Watson is training

That meas our Watson assistant is training and once that message goes we can test our chatbot by sending user examples we have set as well as any related random messages we haven't trained our bot with.

Once message is entered our chatbot should identify the intent of the message and display it under the message

The same way create intents for #thank_you and #goodbyes

NOTICE how I wrote thank_you because intent name cannot have spaces in it

Since we are making a chatbot assistant for a flower shop we would need more intents than just greetings and goodbyes

Create intents for #flower_suggestions and #delivery_info

If a user asks for flower suggestions or flower recommendations then #flower_suggestions intent will be avoked and if asks for delivery information the #delivery_info intent will be activated

Some user examples are

#flower_suggestions

bouquet for girlfriend
flower recommendations
flowers for anniverseries
flowers for a yound lady
flowers for friend
flowers for judges
flowers for marriage function
flowers for my aunt
flowers for my uncle
flowers for your daddy
flowers for your father
flowers for your mother
flowers for your soulmate
flower suggestions
glowers for anniverseries
I'd like to buy some flowers for a sick friend
i want a bouquet for my wife
i want flower for award ceremony
i want flowers for funeral
i want flowers for marriage function
i want flowers for my gf
i want flowers for my loved ones
i want flowers for wedding gift
I want flowers that express sympathy
i want to send flowers to my friend
recommend arrangement of flowers for gettogether
recommend arrangement of flowers for relatives
recommend best flowers for award ceremony
recommend best flowers for award function
recommend best flowers for chief guest welcome
recommend flowers for date
recommend me flowers
suggest best flowers for valentines day
suggest flowers for a date
suggest flowers for award ceremony
suggest flowers for funeral
suggest flowers for girl
suggest flowers for greeting chief guest
suggest flowers for honouring chief guest
suggest flowers for teachers day
suggest flowers for wedding
suggest flowers for wedding function
suggest some nice flowers
what kind of flowers
what should be a good arrangement to give someone when they're retiring
which flower for a birthday
which flower for a funeral
which flower should i buy
which type of flowers are good choice as anniversery gift for your parents
which type of flowers are good choice when meeting someone
which variety of flowers do you have
...

#delivery_info

Do you deliver
do you deliver during holidays
do you deliver on weekends
how do i get my flowers
how much is delivery expenses
how much time in delivery
how will you deliver
is delivery free
what do ou deliver
when do you deliver
when will I get my delivery
when will i get my flowers
when will i receive my flowers
when will you deliver
will i be able to get my flowers on sundays
...

There can be lots and lots more of user examples and even mistakes in spelling from the user side so the more the examples better the assistant gets trained.

Importing Entities

We can actually create a csv file which then can be imported to our watson assistant. In Entities section hover on the first arrow just after the Entities button, it should say something for import

Click on the import button and you will find the format of csv file for importing entities. Create a file like that and then choose the file from your computer

Entities will be imported and watson will be trained with them. Don't forget to check chatbot after training each time.

Creating Dialogs

In the dialogs section of our chatbot tool you should see that two dialogs welcome and anything_else nodes are already there for us with default values

If you try out the chatbot again you should see a welcome response from the chatbot. We will make this response more relating to our florence shop so click on the welcome node and in the Then respond with: section enter your preferred text for welcoming, the text should be very limited to the scope of our chatbot and should not include all the features that our chatbot supports, a nice welcome message would be

Hello. My name is Florence and I am a chatbot. How can I help you? You can ask me about 
flower suggestions or delivery information.

Now try out the chatbot again and it welcomes us with the new message.

If you type in Hello to the chatbot it doesn't respond with anything but we have our intent for greeting ready with lots of examples. So for greeting intent to take action we need to create a dialog node for greeting.

Click on Add node
Name your node for e.g. Greeting
In If bot recognizes: portion enter the entity name
```
If bot recognizes:
#grettings
```

Next enter the response

Then respond with:
Hi I am here to help you. Feel free to ask me for flower recommendations or delivery information.

Close by clicking on cross and try out the chatbot again

Similarly create dialog nodes for Thank You but be sure to put your nodes above anything_else node, you can move your nodes by clicking on the three dots

Once we have our nodes ready we can jump onto more advanced chit-chat dialog flow. If a user asks about flower suggestions for his father our chatbot will recognize the entity but won't respond. For that we have to make a parent node which handles flower suggestions and then child nodes from it which responds to @relationship:father and we need to make our parent node jump to the child node using the options given at end of node.

For recommending same flowers to more than one relationship we can use and or responses

For more relationships create more child nodes with same name Relationship Suggestion and change the condition to the whichever relation you are pointing to and add response accordingly

Integrating Watson Chatbot Into Python

So far we have been making a watson chatbot, there is a lot more that can be improved in our chatbot but we will proceed with that later. In this section we will use the watson python-sdk to communicate with our chatbot using python

In client_app.py we have some imports at the top

import os
import watson_developer_cloud

from dotenv import load_dotenv, find_dotenv

You might be very well aware with os module but the other two modules need to be installed using pip (or any other source)

pip install --upgrade watson-developer-cloud

watson-developer-cloud is the python package to handle the watson python integration.

pip install python-dotenv

dotenv package in python is used to handle the .env file and the env variables inside the file, if you have worked with django you should be aware of the env variables.

Create a .env file and .gitignore file if not exists in the main folder and add .env in .gitignore

Your .env file should have

WATSON_USERNAME=YOUR USERNAME
WATSON_PASSWORD=YOUR PASSWORD
WATSON_WORKSPACEID=YOUR WORKSPACE ID
WATSON_VERSION=YOUR VERSION

Put your credentials from the menu > deploy in the watson chatbot tool

In the docs for python watson client application you can find this in more detail.

Once you have your env file ready you can use the variables with watson_developer_cloud

client_app.py

import os
import watson_developer_cloud

from dotenv import load_dotenv, find_dotenv

load_dotenv(find_dotenv())

service = watson_developer_cloud.AssistantV1(
	username=os.getenv("WATSON_USERNAME"),
	password=os.getenv("WATSON_PASSWORD"),
	version=os.getenv("WATSON_VERSION")
)

workspace_id = os.getenv("WATSON_WORKSPACEID")

Notice we import two methods from dotenv (python-dotenv) package namely load_dotenv and find_dotenv

load_dotenv will load the env variables from the path specified and find_dotenv will locate the .env file from the folder in which the python file is located. We have our .env file in the path as our python file, hence we have successfully loaded our env variables.

Now os module comes in handy, it gives us the variable values from their keys

NOTE: Your env variables should not conflict with the existing system (Computer) variables.

Communicating With Watson

client_app.py

# Start conversation with an empty message
response = service.message(
	workspace_id=workspace_id,
	input={
		'text': ''
	}
).get_result()

We use the service to send a message to our chatbot which is contained inside the input dict with 'text': '' and at last we need to use the get_result() method on the repsonse message so that it is converted into a json serializable object

Now simply print the response and run the file and you should find long json data.

Analyze the json data and print out the useful information like the input and output text


print(f"USER: {response['input']['text']}")
print(f"WATSON: {response['output']['text'][0]}")

Run the file again and you should get the specific information.

Alexmhack / watson_chatbot