Skip to main content
Version: 1.4.0

Build the First Agent

0.1 Create a new agent

Before using the virtual agent to serve you, you need to create and configure the agent.

After logging in to the platform, the first thing you see is the Agent page, where you can add, delete, modify, and check the agent level. Click "CREATE" to create your own agent.

If you cannot see the agent management page, you can refer to Import License.

Note that the language used by the agent is selected here.

Click "Open" to start configuring your newly created agent.

0.2 Create a new FAQ skill

Example scenario: answer the user's question on how to issue an income certificate.

Open "Build-Skills" and click "CREATE". The default skill type is FAQ. You can use all categories first. Then you need to create a new intention on how to issue an income certificate in order to trigger this skill in Question.

Fill in the following information——

Intent Name: How to issue proof of income

Examples: Where to open an proof of income (it is recommended to enter more than ten sentences with the same meaning, preferably more than 20)

Equal to Keyword: Proof of Income

Include keywords: Issuance of income certificate

Finally click "Save". After entering the reply on the new skill page, click "Save".

In this way, you have completed the creation of the skill, and the effective status of the click skill becomes "ON". After clicking the "Train & Publish" agent, you can experience it in the "Test Bot".

Try to ask "Can the Proof of Income be opened for help?", you can see that the agent responded with the corresponding answer. Click "Detail" to view the details of the response.

It is easy to find that, to answer user's question, the agent does not require the question to be exactly the same as the one craeted during Build-Skills step. Agent has a certain flexibility to handle variants of question.

In the scenario mentioned above, if the agent responds to the knowledge points from the user’s question, it is a recall. If the robot’s knowledge points are correctly responded to, it is an accurate recall.

0.3 Create a new dialogue tree skill

0.4 Create a new dialogue flow

Example scenario: Taking a coupon receipt scenario as an example, the user needs to ask the type of consumption selected by the user as a reference for the later distribution of materials.

User: Can I take part in the event and receive coupons?

Customer service: Please choose one of the following three coupons: TV, digital accessories, sweeping robot

User: Digital Accessories

Customer Service: OK

To build such a dialogue requires the use of dialogue flow skills. With this skill, you can freely drag and drop related units on the canvas to more flexibly complete some dialogue scenes with circular and complex circulation situations.

1. New dialog flow skills

On the "Build-Skills" page, click CREATE, select the type for the dialog flow, and the skill name is "Coupon Issuing Activity". After saving, click to switch the effective status to "ON", and then click "Enter".

After entering the process, what you will see is the process editing canvas. The left side is the usable unit. There is a start node on the left of the center of the canvas. The start node means when to enter the task.

Click the "Edit" button of the start node, and you will create a new "How to Get Coupon" intent to trigger, and add examples and keywords as follows:

Intent name: How to receive coupons

Similar statement: Where can I get a coupon (it is recommended to enter more than 10 sentences with the same meaning in the project, preferably more than 20)

Equal Keyword: receive coupon|get coupon

Include keywords:

Click the unit to add the corresponding unit to the canvas. The buttons in the upper right corner of the unit are "Edit" and "Delete".

After the user inquires about the distribution of the learning materials, the agent asks questions. The agent asks the user, and you use the request block.

Click the edit of the request block. The question you might want to ask is "Please choose one of the following three coupons: TV, digital accessories, sweeping robot", and add it to the response box.

The information you will need to collect here is the coupon type, then you use the slot to load the coupon type to be filled in in the future, and create a new slot "coupon type".

The type of slot filling you use is to fill the slot with the entity in the slot fill by entity, click to create a new entity "Activity Product Category", and the entity value is "TV, digital accessories, sweeping robot".

Save the entity, slot, and request block. Here, the query unit can send the user "Please choose one of the following three coupons: TV, digital accessories, sweeping robot", when the user says something When there are related information about "TV, digital accessories, sweeping robot", the key information in the user's speech can be identified and the corresponding entity value can be filled.

Drag the line from the right side of "Start" to the left side of "request block 1" to tell the agent that it needs to flow to "request block 1" after it starts.

After asking about the user's product category selection, you need to send a message to the user. Here, use the "Inform Block" to send the message, add the response as text, and fill in "OK" in the content.

Connect the Request Block and the Inform Block.

Finally, click on "Train & Publish" in the upper right corner. After publishing, you can experience the agent in the Test Bot, and click Reset to clear the session record.

0.5 Fast optimization

In the "Optimize-Label & Review" module, you can quickly conduct agent education and optimize the response effect of the agent. In the Lable page, you can see a lot of user messages. These user messages are recommended by the system that you need to check to confirm the correct response.

For each user message, click the "correct" button behind the intent of Correct Correspondence.

Of course, if you want to directly optimize the wrong answer, you can check the "Add those Wrong & Miss recall intent to intent trigger automatically." option in the settings.

If you want to confirm it again, you can view the relevant content in the Review page, select the relevant marked sentence after filtering, and click "Add to intent trigger".