Graphlogic.ai documentation

How to create a bot

Creating a Bot with Speech Recognition

In this step-by-step instruction, you will learn how to create a small

@Agent⁠

for an online store. The

@Agent⁠

will answer frequently asked questions regarding online ordering, product delivery, etc.

⁠

Creating Intents

First of all, you need to create

@Intent⁠

On the

@Platform⁠

@Intent⁠

s are represented as sets of

@Training Phrase⁠

We suggest that you use

this table⁠

to create

@Intent⁠

s for your test bot.

@Platform⁠

or log in to an existing account using the link

https://app.chatbots.gms-worldwide.ch/platform⁠

. You will be directed to the

@Project⁠

s tab. Learn more about registration and authorization:

Registration and authorization⁠

⁠

Let's create

@Intent⁠

s. To do this, go to the

@Resourses Tab⁠

by clicking Resources on the upper panel.

⁠

You will see a section with

@Intent⁠

s. Learn more about the tab:

Intents tab⁠

⁠

To add a new

@Intent⁠

, click +Create new.

⁠

Let's add an

@Intent⁠

about discounts, sales, and special offers.

In the Name field, enter the name of the

@Intent⁠

“Discounts and Sales”. The names of the

@Intent⁠

s should be concise and clearly reflect their essence. You can find all the names in the attached

table⁠

⁠

Add the

@Training Phrase⁠

s for the

@Intent⁠

in the Requests field. The

@Training Phrase⁠

s for each

@Intent⁠

are also included in the attached

table⁠

. You can copy them from the table and paste them into the field.

⁠

When creating the

@Training Phrase⁠

s, consider the various ways the

@Bot User⁠

may express a specific intention, or

@Intent⁠

, and make the

@Training Phrase⁠

s as diverse as possible.

Important: To ensure the most successful recognition, the
@Training Phrase⁠
s should contain at least 10
@Training Phrase⁠
s for each
@Intent⁠
.

After adding the

@Intent⁠

name and

@Training Phrase⁠

s, save the

@Intent⁠

by clicking the Create button.

⁠

In addition to the main

@Intent⁠

s, it is also necessary to add

@Intent⁠

s for Small Talk. Small Talk represents topics that are unrelated to the main theme of the

@Agent⁠

@Intent⁠

s. This includes, for example, gratitude, greetings, farewells. Small Talk makes communication with the

@Agent⁠

more lively and pleasant.

⁠

@Intent⁠

s for Small Talk can also be found in the

table⁠

Similarly, add all the other

@Intent⁠

s from the

table⁠

. Next, let’s proceed to building the actual

@Script⁠

of the

@Agent⁠

Creating an Agent Script

The

@Intent⁠

s are ready, now you need to create the

@Agent⁠

and its

@Script⁠

Go to

@Projects Tab⁠

by clicking Dashboard button on the upper panel.

⁠

Create a new

@Agent⁠

, as described in the article

Creating a Simple Survey Bot⁠

Go to

@BotBuilder⁠

by clicking the DESIGN button.

⁠

Add an

@NLU⁠

@Slot⁠

by clicking the + button after the

@Start⁠

@Slot⁠

and selecting the desired type from the dropdown list. The

@NLU⁠

@Slot⁠

analyzes the semantic content of the

@Bot user’s Message⁠

and determines which

@Intent⁠

it belongs to.

⁠

In the NAME field, enter the name of the

@Slot⁠

and set the CONFIDENCE THRESHOLD value to 30.

⁠

The

@Confidence Threshold⁠

value determines the

@Agent⁠

's recognition of the

@Bot user’s Message⁠

. Each

@Request⁠

is assigned a confidence value (ranging from 0 to 1), indicating the

@Agent⁠

's certainty regarding the association of the

@Request⁠

with a specific

@Intent⁠

If the confidence value, when converted to a percentage, is higher than the set

@Confidence Threshold⁠

, the

@Agent⁠

proceeds to the branch following the corresponding

@Intent⁠

. If the confidence value is lower than the

@Confidence Threshold⁠

, the

@Agent⁠

goes to the

@fallback⁠

@Subslot⁠

. In this case the

@Bot User⁠

should be prompted to rephrase the question.

After the

@NLU⁠

@Slot⁠

, you need to add

@Intent⁠

When creating the

@NLU⁠

@Slot⁠

, two

@intent⁠

@Subslot⁠

s and one

@fallback⁠

@Subslot⁠

are automatically created.

⁠

Replace the

@Intent⁠

s in these

@intent⁠

@Subslot⁠

s with the desired ones by double-clicking on them and selecting a new

@Intent⁠

from the dropdown list in the Intent field.

⁠

You can clear the Name field so that, after saving the

@Slot⁠

, the name of the

@Intent⁠

will be displayed in the

@Script Tree⁠

, simplifying navigation through the

@Agent⁠

@Script⁠

⁠

Save the

@Subslot⁠

by clicking the Save button.

⁠

After replacing the

@Intent⁠

s in the automatically created

@Subslot⁠

s, create a new

@Subslot⁠

by clicking the + button.

⁠

Select the type of the

@Subslot⁠

from the dropdown list.

⁠

Select the desired

@Intent⁠

from the dropdown list in the Intent field and save the

@Subslot⁠

by clicking the Create button.

⁠

Similarly, add all the previously created

@Intent⁠

s to the

@Agent⁠

@Script⁠

As a result, the

@Script⁠

appear as follows:

⁠

Next, after each

@Intent⁠

, you need to add the

@Agent⁠

’s response to the

@Intent⁠

To do this, after each of the

@intent⁠

@Subslot⁠

s, you need to add

@Text⁠

@Slot⁠

s by clicking + and selecting the required

@Slot⁠

type from the drop-down list.

⁠

Add the corresponding answer from the

table⁠

to the TEXT LIST field:

⁠

We recommend leaving the NAME field empty. So, when looking at the

@Script Tree⁠

, you can see the contents of the

@Text⁠

@Slot⁠

If the text response to the

@Intent⁠

turns out to be long (longer than 4 lines), we recommend dividing it into several

@Slot⁠

s, focusing on the semantic content. This will make the message easier to read. In our example, the response to the Order Tracking

@Intent⁠

contains 7 lines.

Divide it into two

@Text⁠

@Slot⁠

s, adding one

@Text⁠

@Slot⁠

after the other.

⁠

Do the same with the Pickup

@Intent⁠

⁠

If desired, you can add several text response options so that the

@Agent⁠

responds differently each time (the response option is selected randomly by the

@Agent⁠

). This way the

@Agent⁠

will be more “alive” and animated.

Each new answer option is entered into the TEXT LIST field by pressing Enter twice. Add answer options for the

@fallback⁠

@Subslot⁠

: “I did not understand you. Please ask your question in different words.” and “Unfortunately, I don't understand what you mean. Ask your question differently.”

⁠

In the same way, add several answer options for Small Talk

@Intent⁠

After adding responses to

@Intent⁠

s, you need to add

@Wait For Reaction⁠

@Slot⁠

s. This

@Slot⁠

is designed to wait for the

@Bot User⁠

to send a message (

Waiting for the Bot user's message | Wait for reaction slot⁠

). Thus, after receiving an answer to the

@Intent⁠

, the

@Bot User⁠

can ask another question, the

@Conversation⁠

will return to the start of the

@Script⁠

, and the User’s question will be processed by the NLU Slot.

⁠

Agent training and testing

After creating the

@Agent⁠

@Script⁠

, you need to train the

@Agent⁠

and test its

@Script⁠

To train the

@Agent⁠

, click the TRAIN button located in the upper right corner of

@BotBuilder⁠

. It will initiate the

@Script Validation⁠

process. If the

@Script⁠

meets the

requirements⁠

, the

@Agent⁠

will be trained. Otherwise, the

@Script Validation⁠

will fail and an error text will appear in the lower right corner. Read more:

Requirements for successful Agent training⁠

⁠

If errors are detected in the

@Script⁠

, identify problem areas using the guidelines outlined in the article

Requirements for successful Agent training⁠

and fix the errors. After this, train the

@Agent⁠

again.

After successful

@Agent⁠

Training, open the

@Debug Widget⁠

by clicking the Debug button.

⁠

Ask the agent some questions in accordance with the created

@Intent⁠

s. Try formulating the questions in different ways using synonyms.

During testing in the

@Script⁠

, the

@Slot⁠

in which we are currently located is highlighted. This allows us to track which specific

@Intent⁠

the

@Agent⁠

attributed the

@Bot user’s Message⁠

to.

For the most effective testing, we recommend reading the article

Testing the Agent script⁠

To choose an appropriate Confidence threshold, refer to the article

Testing and improving Agent recognition: selection of Confidence threshold⁠

You can retrain the

@Agent⁠

directly from the Analytics tab. More details:

Additional training of the Agent from the Analytics tab⁠

After editing the

@Training Phrase⁠

s of the

@Intent⁠

s, don’t forget to train the

@Agent⁠

again.

Communicate with the

@Agent⁠

, going through all the steps of the

@Script⁠

, while checking that there are no errors in the texts and the logic of the

@Script⁠

is correct.

If you find errors, fix them, remembering to retrain the

@Agent⁠

after each change so that it takes effect. If everything is correct, congratulations, your

@Agent⁠

is ready!

Creating Intents

Creating an Agent Script

Agent training and testing

Want to print your doc?
This is not the way.

Try clicking the ⋯ next to your doc name or using a keyboard shortcut (

CtrlP

) instead.