The Chatbot Implementation:
Use IBM Watson Assistant Service in IBM Cloud for Chatbot Implementation
Date: 04/05/2022
Introduction:
“It is quiet early for your case but just eat right, exercise often, and take your medication, you
will be fine.” The above was a medical advice given to me by the Doctor, sometime in February
of this year. It was like a fight, trying to stay healthy, to bring sugar level to a normal range. I am
just 42 years old with signs and symptoms of diabetes. This is no different from the cases of
many people in America, and the whole world. According to the Center for Disease Control and
Prevention (CDC), over hundred million people in America are, already, diagnosed with diabetes
or prediabetes while some are still undiagnosed. This is gradual, but increasingly becoming
prevalence. Most people are not aware of their health condition when it comes to diabetes and
sugar level monitoring. While many do not have the time or resources to properly take care of
their health. The immediate need for a poor person is to satisfy hunger with any available food.
Unfortunately, majority of the affordable meals are not healthy, they are either filled with
calories or fat. Continuous consumption of food with high calories (rich in sugar) will result in
high sugar level, which leads to diabetes. The process of digestion involves breaking down of
food into sugar (also called glucose). Eventually, these glucoses are released into your
bloodstream, thereby, leaving the sugar level in the blood either low, normal, high, or in some
extreme cases, ketoacidosis. The pancreas helps in regulating the sugar level, by releasing them
to used as energy, when they are high. “If you have diabetes, your body either doesn’t make
enough insulin or can’t use the insulin it makes as well as it should” (CDC). Early detection and
awareness will help anyone to manage the sugar in-take and practice healthy living like
exercising more. This is where Mr. Bot Diabetes (Chatbot) comes in to play. Figure 1 shows
clearly that diabetes diagnoses are on the rise, having compared the county-level distribution of
diagnosed diabetes prevalence among US adults aged 20 years or older, 2004, 2008, and 2016.
Figure 1: County-Level Distribution of Diagnosed Diabetes Prevalence Among US Adults Aged
20 Years or Older, 2004, 2008, and 2016
Hence, the chatbot is an expert on sugar level, an assistant, who will welcome the user (anyone)
with the intention to advise its user on blood sugar level monitoring. It will start off with an
introduction, saying
its name (Mr. Bot Diabetes), explaining to the user what it can do, and inquire for a name from its
user, so that it can address user by name. Moving forward, the chatbot will find out if the user
want have concerns about sugar level or just want some information of sugar level management. If
the user indicates interest in talking about sugar level, the chatbot will find out if its user already
knows the reading (sugar level – number) and the time when the number was gotten. The time
will be either before meal, during meal, 2 hours after meal, or at bedtime. If the user knows the
number and time, the chatbot will advise its user according to the number of the reading and the
time it was gotten. On the other hand, if the user does not know the blood sugar reading, the
chatbot will ask its user the symptoms noticed. Different symptoms associated with different
sugar level, low, high, or normal. However, the chatbot will advise based on the signs and
symptoms mentioned by its user. Similarly, if the user declines interest in talking about sugar
level, the chatbot will find out if user wants some information on blood sugar monitoring. The
flowchart below is the dialog flow diagram.
Figure 2: The Dialog Flow Diagram
Dialog Implementation:
Having completed the flowchart above, Watson Assistant service was used to build the chatbot.
The dialog implementation utilizes the user entities and 3 of the system entities, intents, handlers,
context variable and slots, the dialog nodes and flow. The purpose of the chatbot is to chat with
users and provide advice and/or information to them. The developmental steps begin by creating
the skill, using the dialog
skill template. The skill contains the training that will respond to the user intent. Provide the
name of the skill. In this case, the name of my skill is Mr. Bot Diabetes link: https://web-
chat.global.assistant.watson.cloud.ibm.com/preview.html?region=us-
south&integrationID=3781c303- 9c60-4a43-8c4e-
2e66894507d5&serviceInstanceID=d8e91774-4245-4873-91bc-e6a97c10c379 Then, followed
by the creation of the intents, entities, and the dialog nodes and child nodes.
An intent is a collection of user’s possible statement. This will help the assistant to easily
understand what the user has in mind. While on the other hand, entities are like keywords with
their synonyms. Within my skill, I created symptoms entity to collect information on the sugar
level signs and symptoms. The dialog nodes (or child nodes) are used to guide the discussion. By
default, assistant starts with “Welcome” and “Anything_Else” nodes. Different nodes and child
nodes are added based on the channels of the discussion. The higher-level topics are treated at
node level, while details are taken care of at the child node level. The dialog uses intents, entities,
and context variables to decide the interactions. The slots are used when multiple entries are
expected. Slots will receive user’s response and create context variable to store it. Within a slot,
assistant will have variety responses in case the user is not responding as expected. Having said
that, context variable is a placeholder within a slot. Context variables are called in a dialog.
In the development of Mr. Bot, I created $name to hole a place for the name user will supply.
Dialog Scenarios:
The following scenarios will be discussed
• User has concern about blood sugar level = YES, and knows the reading of sugar
level = YES:
In this dialog scenario, once the assistant is lunched, Mr. Bot will welcome its user,
explain to the user its mission, and find out user’s name, then wait for reply. For example,
“Hello, my name is Mr. Bot Diabetes. I will talk with you about blood sugar levels. May
I know your name?” or “Greetings, I am an expert on blood sugar. I will chat with you
on the topic. What is your name?” After the user supplies a name for the chat, the
assistant will address the user by name and asks to know if the user has concern about
blood sugar. If the user selects yes from the options available, then assistant will ask to
know if the user already knows the sugar level. If yes is selected again, from the options,
the assistant will prompt the user to enter the number of the sugar level reading. At this
time, the @Sys-number entity will be used to check if number value is entered, and the
number will be saved in the $sugar_level_reading context variable. Assistant will move
to find out when the number was gotten e.g., before meal, during meal, after meal, 2 hours
after meal, at bedtime etc. These intents have their synonyms to accommodate different
varieties of answers for different users. When a valid blood sugar reading and the time it
was gotten are received, assistant will check through the eight conditions that were to be
met before offering its findings and recommendations. For example, a normal blood sugar
range before meal is between 80 and 130. So, if user’s input is withing 80 and 130, and the
time is in the morning before meal, assistant response (advice) will be “Your blood sugar
level is NORMAL. Keep eating healthy and exercise more to maintain a normal blood
sugar level.” Other conditions will check for different blood sugar levels and time to
decide whether to categorize the level as “LOW” or “HIGH”, with corresponding
recommendations. Figure 3 below shows the Try-it (testing) screen page.
Figure 3: Try-it Page for Concern = Yes and Know Blood Sugar Level = Yes
• User has concern about blood sugar level = YES, and knows the reading of sugar
level = NO:
Similarly, if the user selects yes to the question, “Hi $name, do you have concerns about
blood sugar levels?”, but selects no to “So, $name, do you know your blood sugar level
(number)?”, the assistant will prompt the user a different question to find out what
symptoms the user has noticed. Again, different symptoms associated with different sugar
levels. With the symptoms the
user supplies, assistant will match it with the corresponding advice. See the figure
below for the Try-it screen page.
Figure 4: Try-it Page for Concern = Yes and Know Blood Sugar Level = No
• User has concern about blood sugar level = YES, and knows the reading of sugar
level = EXIT:
While in this scenario, having asked about the sugar level reading, the user may opt
out of the conversation. The assistant will jump to goodbye node, thank the user with
response like,
“Thanks $name for using this service. Have a nice day. Eat right and stay healthy”
and return to the main menu. The figure below shows the scenario in a Try-it screen page.
Figure 5: Try-it Page for Concern = Yes and Know Blood Sugar Level = Exit
• User has concern about blood sugar level = NO:
However, if user does not have concern about blood sugar, that is the user shows no
interest at the beginning of the dialog, assistant will want to know if the user is interested
in more information about diabetes and prompt the user with options to select from
“Information on Diabetes” or “Exit”. If the user chooses to exit, the assistant will jump to
goodbye node and appreciate the
user for using the services. But if user goes for information on diabetes, the assistant will
prompt the user CDC link to redirect the user to read for more information on diabetes
and jump to the main menu. See the Try-it screen page below.
Figure 6: Try-it Page for Concern = No
• User has concern about blood sugar level = EXIT:
Finally, if the user gives name but decides to quit, the assistant will jump to goodbye
node to end the discussion. See the Try-it screen page below.
Figure 7: Try-it Page for Concern = Exit
Integration with Watson Discovery Service:
Watson Discovery services will serve as a repository. The purpose of the repository is to
house all the documents collected from different users. The Web Crawl connector was used to
build the repository. CDC website was provided to synchronize with my search skill.
Figure 8: The Assistant Created
Figure 9: The Overview of the Assistant
After deployment, the preview was tested. The figure below shows the preview page.
Figure 10: The Preview Page
Conversation Data Analysis (2-days experiment):
The link was shared with 5 friends and 2 days conversation was captured. Total of 29
conversations were captured and analyzed. The figure below shows the dashboard and log of the
conversation.
Figure 11: The Dashboard of the Conversation
Figure 12: The Conversation Log
From the analysis, total active users are 29, between 4/4/2022 and 4/5/2022. See the figure
below. Based on the name provided to the chatbot, an average of 7 names were part of the
conversation. There are 111
messages which gives an average of 3.96 messages per conversation, and an average of 4
conversation per user.
Figure 13: The Analysis of the Conversations
The last part of the analysis page shows the top intents and entities used.
Figure 14: The Top Intents and Entities
Chatbot Enhancement:
Speech-to-Text, Text-to-Speech, Language Translator, and Tone Analyzer can be integrated to
in the assistant to enhance user experience. For a user who is blind, Speech-to-Text can help
such visual- impaired person to dialog with the assistant, thereby, giving opportunity to anyone
who cannot write for any reason. Also, if Speech-to-Text is integrated, there will be need to
bring in the Text-to-Speech. For the case of a visually impaired, reading assistant’s greetings
and instruction will pose as another challenge. Therefore, Text-to-Speech will be necessary to
come onboard. Since Mr. Bot was developed using English language, Language Translator will
be integrated to be able to serve other languages not English. It will be a good thing to monitor
and analyze the tone of users. Therefore, Tone Analyzer will be good to integrate to the
assistant. This will help to encourage users if the problem of sugar level is taking a toll on the
user.
Diabetes diagnoses is trending high within and outside the United States, and has cut across all
age groups and race, therefore, I would consider using all messaging platforms available, to
reach out to as many as possible.
Conclusion:
The experience worth the time and exercise. Mr. Bot Diabetes was successfully executed
different user’s intent and provide advice and recommendations accordingly. In fact, I did not
prepare my sister beforehand, and when she got recommendation from Mr. Bot, she became
worried and reached out to me. See the screenshot of my WhatsApp chat with my sister.
Figure 15: The WhatsApp Chat Between My Sister and I
It was that good. So, for different dialog scenarios when the user completes the conversation
with Mr. Bot, talking about blood sugar level and monitoring, the assistant will respond with
one of the following recommendations.
“Your blood sugar level is LOW. I recommend you taking 4oz of orange juice or any
regular cola drink”
“Your blood sugar level is NORMAL. Keep eating healthy and exercise more to maintain a
normal blood sugar level.”
“Your blood sugar level is HIGH. I recommend you talk to your doctor about how to keep
your blood sugar levels within the normal range. Consider using an over-the-counter ketone
test kit to check your urine for ketones and call your doctor if your ketones are high. Quit
eating or drinking any food that can affect blood sugar levels.”
“This number may be inaccurate due to the sugar content in the food since you check it during
meal. Best time to get an accurate reading will be 2 hours or more after meal or before meal.”
However, the chatbot implementation comes with few challenges that I can classify as personal.
The development is more of software development that data analysis. Therefore, experience in
coding and most logic behind coding are required. Again, my instructor already pointed out the
number type. It will be perfect if one can program the assistant to receive any number but do the
necessary conversions internally. For example, inches to feet, Celsius to Fahrenheit, as so on.
Finally, Mr. Bot Diabetes has come to assist human being with sugar level management and
refer to the user to a live person, physicians. In fact, it will save the user the cost for frequenting
the doctor’s office just the get advice on blood sugar monitoring.
