This chapter will explain: what is FAQ and how to build FAQ skills on the platform.
Definition of FAQ
FAQ is used in a clear single-round inquiry scenario. A FAQ consists of user questions and reply answers. The robot recognizes the user’s question and gives the corresponding reply answer.
For example, there is a question asking "What should I do if my baby has a fever".
The question may include "the baby has a fever", "my baby has a high fever" and "my baby has a fever, how can I cool down" and so on.
The corresponding answer may be "If the baby's body temperature does not exceed 38 degrees, it is recommended that you use a cold towel to put on your forehead for physical cooling; if it exceeds 38 degrees, it is recommended to go to a regular hospital for treatment."
The above-mentioned questions and answers on "What should I do if my baby has a fever" constitute a FAQ.
Similarly, government policies, corporate regulations, consultations, and shop activity rules are all FAQ scenarios.
When a trainer creates a FAQ skill, only two items need to be configured on the platform: -User question: the "intent to trigger the skill", the question asked by the user can be regarded as an intent -Reply answer: After triggering this intent, the agent's reply response is generally one or more answers.
- On the skill management page, click the "CREATE" button and select Type as FAQ to create a new FAQ:
- First select an intent as the user question. If there is no suitable intent, you can click CREATE to create one:
At present, the platform provides two ways to trigger FAQ intentions: trigger by keywords, and trigger based on semantic similarity matching. -Keywords: perform character matching on the keywords that appear in the user's message. If the matching is successful, it means that the user has triggered this intent. -Similarity matching: The platform performs semantic matching on the similarity between user messages and knowledge points. If the similarity score reaches threshold, it means that the user has triggered this intent. For more ways to use intents, please see the Intents page
- After selecting the intent, you can click the "Preview detail" button to quickly view the content of the intent, or click the "Edit Intent" button to modify the intent directly.
- Next, complete the configuration of the reply answer corresponding to the user question. Currently, three answer types: text, picture and audio are supported
|Reply Type||Description and Restrictions|
|Picture||2MB, support .png .jpeg .gif format|
|Voice||2MB, support mp3 wav format|
|File||1GB, multiple formats|
|Video||1GB, supporting MP4|
|Rich text||support common rich text|
-The format style supported by specific channels depends on the corresponding channel configuration
-A single knowledge point can add up to 10 reply answers
-Rich text is recommended to be pasted after editing in Word
The effects displayed on test bot are as follows:
- Knowledge Q&A supports the configuration of the robot's reply method, that is, when the question is configured with multiple answers, which strategy is used to reply to the user:
-Reply to all in turn -Reply to all in order will reply to the user in order according to the order of the page when multiple reply answers are configured. -For the questions asked by the user, it may not be possible to get a satisfactory answer through only one reply. You need to configure multiple text and picture answers to answer the user’s question. It is recommended to choose the strategy of replying to all of them in turn. -Random reply -Random reply will randomly select a reply to the user every time when multiple reply answers are configured. -For example, for some knowledge points about greetings and small chats, you may expect that although this knowledge point is also matched, it does not give the same reply every time. This appears to be too dull for the robot. It is recommended to choose a random reply Strategy.
In order to facilitate the construction of the knowledge base, Wulai will not force you to add reply answers when saving knowledge points. However, in order to ensure that the robot can normally respond to the questions raised by the user, it is recommended to add answer content to each effective knowledge point.
- Under the configured answers, click "set relevant intent manually", search for the name of the intent to be associated and click OK. When the answer is replies to the user by the robot, the associated intent will be displayed below the answer for the user to continue to ask questions to the robot by clicking.
- The knowledge quiz has been set up, click save and it will be displayed in the skill list. By default, the newly created skill is in the "OFF" state, set it to "ON" and train and release it to experience the effect.
Optimized FAQ library structure
Avoid corpus redundancy