Skip to main content
Version: 1.4.0


This chapter will explain: strategy definition and how to configure the agent's response strategy on the platform

Idle waiting time

The user may not reply to the robot's questions immediately in a real conversation. At this point, you may want the robot to stop here and wait for the user. If the user replies after a period of time, the dialogue process can continue. If the user never replies, the robot ends the conversation after a period of time.

In order to make the robot have a "pause" intention to wait for some events, the platform supports the idle waiting mechanism. Situations where the mechanism can be used include:

  • After the robot inquired, the user did not respond in time.
  • Other intentions interrupt the current process. The platform allows the current dialogue process to be interrupted by another intention, and then run "another intention". For example, the user inserts a question and answer in the dialog tree

In either case, the current intention is "suspended". Then, starting from the starting point of timing:

  • If you return to the current intention within the idle waiting time, the process will start from the original "suspended" position and continue to execute onward.
  • If the idle waiting time is exceeded, then
    • The process corresponding to the current intention will be suspended;
    • The previously collected information will be cleared;
    • The user message must match the intention trigger to return to the intention again;
    • The dialogue will begin at the beginning of the intention process.


FallBack: When the agent thinks that it cannot give an accurate answer or recognizes no intention, it will go straight to the bottom

FallBack Score The intent that is most similar to the user’s sentence is called the Top1 recall intention. If the similarity score of Top1 recall is lower than FallBack Score, the agent will think that it cannot match the correct intention, so it will choose to reply from the bottom.

This score is called the fallBack score.

FallBack Action is the words that the agent returns to the user when it botches.

For example, a common bottom reply may be set to: "Sorry, I am not sure about this question. I am still learning, can you ask me questions in another way?"

Platform operation

  1. Enter the edit page from the menu "Build-Strategies" The system presets the fallBack score and fallBack action, which can also be modified according to the real situation