Customize AI Chatbot¶
A chatbot creator can use Juji Studio to customize an AI chatbot in three areas:
Customize Main Chatflow Every AI chatbot on Juji is driven by a main chat flow, the conversation agenda that threads one or more topics to drive a conversation forward. As shown below, a chatbot creator can add, remove, and edit a topic on a main chat flow.
Customize Q&A and Fallback AI chatbots should be able to answer user inquiries or respond to user comments any time during a chat. To enable such capabilities, a chatbot creator can add, remove, and edit a Q&A list and fallback handling as shown below.
Customize Chatbot Persona and Beyond A chatbot creator may also want to customize the chatbot persona as well as other chat-related settings, such as controlling the pace of a chat.
Customize Main Chat Flow¶
As shown above, an AI chatbot created based on a template already contains a default chat flow, which consists of one or more topic blocks shown on the left side of the screen. Each topic block often represents a Juji built-in mini dialog that enables a chatbot to automatically carry on a multi-turn conversation with users.
A chat flow is often defined for a chatbot to achieve specific goals. For example, a typical AI sales helper chatbot is associated with a chat flow that asks a user several questions and then use the gathered information to offer personalized services, such as making product recommendations.
Since each chatbot is made for its own missions, you can easily customize a chatbot by altering the chat flow or any topic blocks in the flow.
Add a Topic¶
One can add a topic to an existing chat flow in one of the two ways:
Click on the round green
+button below the
Welcomecard to add a topic at the beginning of a chat flow. You can also use the same green
+button above the
Wrap-upcard to add a topic at the end of a chat flow.
Mouse hover any topic card (e.g.,
How are you doing?) and then select the round green
+button to add a topic after the current topic card.
As shown above, Juji currently supports two types of topics:
This is for a chatbot to ask a user a question or perform a certain action. The chatbot must wait for a user to respond to such a request before moving the conversation forward. Currently, Juji supports five types of chatbot request:
This is for a chatbot to send a message to a user without waiting for a user's response. Currently, Juji supports three types of chatbot message.
IMPORTANT TIP If you intend to have your chatbot wait for a user input and respond to it before moving on, make sure you choose
Make a Request. Otherwise, your chatbot simply ignores any user input even if the message is worded like a question. As the example shown below, T6 will not wait for a user's input but T5 will. Note the chat icon appearing on T5, indicating T5 is a "true" question.
To decide which chatbot request to use, please check out this design tip.
Edit a Topic¶
Once a topic is added, one can then edit the topic. See the section below on how to edit a specific topic. One can edit a topic in three sections:
- Chatbot Content Use this section to customize the actual chatbot message(s) that users will see during a chat. Please refer to examples and best practices for defining the content of a topic.
- Chatbot Actions Use this section to customize chatbot behavior under different conditions. The green "Customization" button shown below can be used to add a group of chatbot actions under one condition (trigger). Please refer to examples and best practices for defining specific conditions and corresponding custom chatbot actions.
- Topic Settings Use this pop-up window to customize one or more topic-specific parameters, such as the delayed start of this topic and required user response length. Please refer to best practices to tune these parameters.
Delete a Topic¶
To delete a topic, mouse hover the corresponding topic card listed in the left topic panel. Then click on the red trash can icon to delete the topic. A deletion is permanent and cannot be undone.
Clone a Topic¶
To clone a topic, mouse over the corresponding topic card listed in the left topic panel. Then click on the green clone icon to clone the topic. The cloned topic will retain all behavior of the original topic.
Although the chat flow is presented a list of topics, but the inter-connected topics actually form a graph. When the complexity of the chat flow grows, the need to see the relationship among topics grows also. Juji Studio provides two convenient tools for users to visualize and examine the topic graph. The two icons on top of the topic list leads to a text and a graph view of the topic graph.
Topic Text List
Click on the Topic Text List icon, you will see all the topics represented in a textual format in a new browser tab. You can copy or download these text for detailed analysis.
Click on the Topic Graph icon, you will see all the topics represented in a graph visualization in a new browser tab. Each node in the graph represents a topic. You can drag the nodes around, pan and zoom in/out the graph. Click Reset button to change the graph layout. Click Center button to re-center the viewpoint. Click Legend button to toggle on/off the legend of the graph. You can also change how long you want the node label to be by choosing a Label Size.
When you click on one of the graph nodes, not only the node turns yellow, the corresponding topic editor in the design view is also selected and open. So you can use the topic graph as a convenient topic navigator for editing topics.
Here is how to do it: first drag out the topic graph tab as a window, then lay this window side by side with the design view, or lay it top and down with the design view. The idea is to show both windows at the same time, so you can use the topic graph to navigate between topics for editing.
Customize Q&A and Fallback¶
In addition to driving a conversation based on the main chat flow, an AI chatbot must handle user questions or side-talking raised any time during a chat. To enable a chatbot to do so, Juji allows the customization of Q&A and fallback. Such customization enables Juji chatbots to answer user free-text questions, handle user digressions, and even chitchat socially, anytime during a conversation.
Handle Free-Text Q&As¶
Q&A Board, one can support custom, natural-language Q&As,
monitor, and manage unanswered user questions encountered during a
chat. Below is an example of a Q&A board:
On this page, one can enter a custom Q&A pair directly in the table
(using the green
+ button to add) or upload a CSV file
containing a Q&A list. The above screenshot also shows that there is
one unanswered user question (indicated by the red badge displayed
next to the
Q&A Board). Juji automatically detects unanswered user
questions and displays such questions on the Q&A dashboard.
IMPORTANT TIP If you decide to put all your Q&As in a CSV file and then upload, download the CSV file first as it contains the required file format to fill in your Q&As.
As shown below, each template includes four columns:
All four columns must be preserved, so is their order. Although the
first column (
ID) and the last column (
Comment) can be
left empty, the empty columns must still exist and cannot be deleted.
ID You can use this column to give your Q&A pair a name/label. If you don't provide an ID, Juji will automatically generate one. This label is especially useful to group all varied question expressions into the same question. This way you don't need to duplicate the answer for each varied expression. For example, rows 2-3 and rows 5-6 represent two sets of Q&As. Moreover, they also share all the alternative answers (see below).
Question This column holds questions and their varied expressions.
Answer This column holds answers. You can enter alternative answers to the same question. This will make your chatbot sound more intelligent. When there are alternative answers, a chatbot randomly chooses one to use when answering a matched question.
Comment This column holds your comments, which is optional.
Support Social Chitchat¶
In addition to answering user questions, you may want your chatbot to
support social chitchat, e.g., responding to user casual comments. Similar
to handling Q&As, you can enter user comments (e.g.,
you are really
smart) and corresponding chatbot responses (e.g.,
Thank you, I'm
flattered). During a chat, whenever users make matched comments, your
chatbot can then respond to them.
Below shows an example of custom social chitchat. Whenever a user
texts expressions similar to those indicated under ID
(B8-B10), the chatbot will tell a joke by randomly picking one from
cells C8 and C9:
IMPORTANT TIP Juji already provides rich, built-in social chitchats. We recommend that you observe the behavior of your chatbot first before adding any custom user comment-chatbot response pairs.
Below is a very short video that shows how to create or update a Q&A list:
Customize Chatbot Settings¶
In addition to customizing the main chat flow and the Q&A list, one can also customize various chatbot settings, including the name of the project.
You can edit the
name of the project by mouse hovering the project
name and then click on the
To personalize your chatbot, you can choose a stock persona or define a custom persona.
To define a custom persona, click on the
Custom persona icon as shown below. You can then upload your persona photo, give your custom persona a name, and enter a short bio. Once you are done, click on the green check mark to save it.
You can also customer other chatbot parameters, such as setting a
conversation pace and
user response requirement as shown below.
Conversation pace It controls how fast a chatbot transitions from one topic to the next.
Refresh duration It controls how frequently a chatbot auto starts itself from the beginning. In certain situations, you wish a chatbot to start more frequently (e.g., an e-commerce chatbot) than others (e.g., an interview chatbot) to make sure that users can accomplish their tasks without starting over again.
Minimal response length It controls the minimal number of words a user must provide to a chatbot request. This parameter sets the default for all questions and it can also be customized for a specific chabot question/request.
- Fallback handling This allows you to control what kind of fallback handling is desired for your chatbot. Juji provides a rich list of fallback handling functions to manage user input that is outside of your chatbot script. From our experience, users often deviate from your chatbot script and start a side talkng. These fallback handling functions enable your chatbot to handle user-initiated side talking gracefully and also keep the chat on track.
In this section, we use a set of concrete examples to show how to customize a chatbot. In particular, we show how to customize a chatbot's content and its behavior.
Customizing Chatbot Content¶
Here we use a set of examples to show how to define the content of different topics and how the content would appear in a chat on the web or Facebook Messenger.
Plain Text Chatbot Message
This is the simplest chatbot message that contains only plain text.
Add a function Use the green function button (
fx) below the text bubble to add a function, such as
(user-first-name)to personalize the chatbot message.
Add a link Use the green
linkbutton to add a URL (
Add a paraphrase Use the green
+button on the right of the text bubble to add paraphrases of the message (see the screenshot below). The paraphrases will be selected by your chatbot randomly to send during a chat. See best practices for good uses of paraphrases.
You can always use the green
eye icon located at the top-right
corner of the text bubble to preview the above message in a chat:
FB Media Card
For a FB Messenger chatbot, one can define one or more FB Media cards to display a combination of text, image, and URL links. As shown below, each card contains:
Title (required field)
Image (optional image URL)
URL Buttons (up to 3 links per card)
Currently one can define up to three FB Media cards. In a Facebook Messenger window, these cards will show up in a carousel.
Note that FB Media cards will not show up if the chatbot is deployed on a website.
Web Media Card
For a website chatbot, one can also define a message with a combination of text, image, and URL. As shown below, the web media card contains:
- Text Message (required)
- Hyper Link (optional)
- Image (optional image URL)
Below is the preview of a web media card with an image:
Perhaps the most used request for an AI chatbot is to pose free-text, open-ended questions during a chat. Such a question can elicit diverse and detailed user responses to gather in-depth insights and better understand users' needs and wants.
Label Typically, a label is the short form ("stem") of a chatbot request. Although it automatically takes the form of the main message, it is good to give a sensible label as it is also used for multiple purposes (see best practices).
Main Message (required) This is the question that users will see in a chat. Instead of asking a question directly, a good message should also contain the context or motivation of the asking.
Re-asking Messages (optional) During a chat, a user may not always respond to a chatbot question. In such a case, a chatbot may want to repeat the question. One can add (green
+) paraphrases of a request and indicate whether such a paraphrase can be used to re-ask the question by checking the
See best practices on how to phrase free-text questions in a conversation.
One of the popular chatbot request is to ask a user to make a choice. Below is an example of a single choice request, also known as radio button question, where a user can select only one choice.
As shown below, a single choice question contains:
Main Message (required)
Option Items (at least one item is required)
In addition to adding regular option items,
Other option can be added to
let a user specify additional text. One can also indicate whether to
treat all the option items as numbers. Numeric choices can be used to
define chatbot conditions involving numeric operations, such as
<. See examples below on defining chatbot conditions.
Depending on where the chatbot is deployed, the look of a choice question may be different.
A choice question displayed in a web-based chatbot:
A choice question displayed in a Facebook Messenger chatbot:
In Facebook Messenger, a user may enter a text message instead of clicking on a choice. In such a case, the chatbot will handle the user text input first. Since a Juji AI chatbot tracks a conversation context, it will repeat the question, if the choice question is a required one.
Below is an example chat where the user asks a question intead of making a color choice. The chatbot answers the user question and then repeats the choice question.
Semantically, a Facebook choice question is essentially the same as a regular choice question except that it takes the form of Facebook buttons in a Facebook Messenger chatbot. This type of request will not be displayed in a web-based chatbot.
As shown below, a Facebook choice question includes:
- Title (required) This is the main asking message.
- Subtitle (optional) This adds more information to the question.
- Button (at least one button is required)
- Image (optional) An image URL can be added with the question.
Here is how a Facebook choice looks like in a Facebook Messenger chatbot:
This special request enables a chatbot to gather opt-in user email in
a Facebook Messenger chatbot. A user must click on the displayed email
to confirm (opt-in) his/her email. A user can also text an alternative
email. To allow users to skip this question, set this question
required in the topic settings.
Customizing Chatbot Actions¶
In a natural conversation, different user behavior should drive different chatbot actions. Juji supports the customization of chatbot behavior in many ways. This section will use a set of concrete examples to show how to easily customize a chatbot behavior based on application needs.
Unlike other chatbot platforms, Juji has a dialog library that contains thousands of built-in mini conversations. When you add a topic that is a free-text request, Juji automatically uses the request label to search its dialog library and retrieve a matched mini conversation for this request.
For example, if you enter
Have you ever worked at a restaurant? as
the request label, Juji will auto retrieve the matched mini
conversation `This is to ask a user yes/no question' (see below)
In this case, the built-in dialog will automatically process a user's positive or negative responses and enable you use the auto-detected expressions to direct the user to different paths. In other words, you don't need to worry about how to define triggers to catch highly diverse users' positive or negative expressions.
In contrast, if the request label is
What's your favorite fruit, the
retrieved mini conversation would be
Ask a user about his/her
favorite thing (see below)
In case that Juji-retrieved mini conversation is not what you want to use, you can always find a different built-in dialog (click on the green search icon as shown above). Below is a screenshot showing the Juji dialog library, where you can browse and use keywords to search for a suitable, built-in dialog to use.
A built-in Juji dialog is very powerful as it helps handle diverse
user behavior automatically. For example, if you wish to collect
users' contact information, the built-in dialog
Ask a user contact
info automatically handles various cases, such as user already opt-in
contact information or potentially erroneous input.
In certain built-in dialogs, Juji automatically creates a set of
built-in attributes to capture processed user input. For example, if a
mini dialog talks about the
weight of something (e.g.,
does your dog weigh), Juji automatically parses user input to the
question and stores the parsed user input into two attributes:
Click on the green
attribute icon to view the attributes associated
with a built-in dialog (see below).
You can also edit the attribute names to avoid potential conflicts. For
example, your chatbot asks two questions regarding
weight, one for
dog and the other for cat. You would want to change the default
attribute names to
respectively to distinguish user answers to these two questions.
Pin a Matched Dialog
As described above, Juji uses the request label you entered to search
and find a matched built-in dialog. This means that changing your
request label will automatically trigger Juji to find a different
matched dialog. To avoid such situations, you can use the green
icon to pin a found dialog (see above). This means that changes in
your request label will no longer trigger Juji to search and
change the matched dialog automatically.
Besides reusing a Juji built-in dialog, you can always customize a built-in dialog. Next we use a set of examples to show how to customize built-in dialogs.
As shown below, each customization block includes two parts:
IF block defines the condition (also known as trigger) under which the custom chatbot actions will be performed.
THEN block contains one or more custom chatbot actions.
Customize IF Conditions
Since a free-text chatbot request may elicit unanticipated, diverse user behavior, currently Juji supports five types of IF conditions to capture user behavior:
is-similar-toIf a user input is similar to one or more sentences or phrases. One can also specify the similarity threshold. By default, the threshold is set at 80%.
contains-keywordsIf a user input contains one or more keywords. Unlike
is-similar-tocondition, this condition requires a precise match with the stem of at least one specified keyword.
matchesIf a user input matches a rule specified in Juji chatbot language.
is-anything-elseIf a user input matches anything else. This is basically the default condition.
One can check on the box next to the
STORE MATCHED INPUT to store the
matched input into a custom attribute.
As shown below, if a user input contains keywords, such as
cheese, a custom attribute
like-protein can be created
to store the matched keyword for each user.
Custom attributes can be used for many purposes, such
as customizing a chatbot message (see below) or indexing audience
information (e.g., showing all users who
Note that for a choice request, since a user's input must be one of the choices, there is just one type of IF condition, the choice made by the user.
To decide which type of trigger (chatbot condition) to use, please check out this design tip.
Customize THEN Actions
As shown below, when a
IF condition is met, one or more custom
chatbot actions can be defined:
Quick AcknowledgementThis defines a simple text message that a chatbot can use to acknowledge a user. This simply sends a message without waiting for a user's response. This means that any follow-up question expression in this message won't be able to elicit any user response (unless it is a reflective question without requiring any user response). If you wish to ask a follow-up question, add a
Follow-up Requestas described below.
Extended ReplyThis defines a full chatbot message, including image and paraphrases, that a chatbot can use to make an extended reply. This defines a full topic - a message topic.
Follow-up RequestThis defines a chatbot request that allows the chatbot to ask a follow-up question. This defines a full topic - a request topic.
Suspend Current Chat
Sometimes, you may want to pause a chatbot a bit before letting it interact with your audience again. You can do so easily following these steps:
Main Chat Flowunder
Click on the
Click on the slider under the welcome message (on the right panel) to suspend a chat. See the screenshot below.
Edit the welcome message accordingly to inform users about the suspension/pause.
IMPORTANT TIP If your chatbot is deployed via a web URL, you may also want to change the greeting message on the web cover page to inform future users about the pause/suspension.
Customize End Topic
One may want to customize a chatbot's ending behavior. For example, if a survey chatbot needs to end a conversational survey at some point. In contrast, a customer service chatbot must hang around to serve customers.
To customize the ending behavior, click on the
topic, the last topic in the left topic panel.
- Then click on the text bubble or the
pencilicon to edit the topic:
Option 1: Make a chatbot stick around
Option 2: End a chatbot
Once a chatbot is ready, you can deploy it onto a website or a Facebook page. Please check out Chatbot Deployment to deploy your AI chatbot.