Quick BI:Prepare Data

Q&A configuration

Before you can use Q-Copilot, you must configure the Q&A settings for the dataset.

Access

You can access the Q&A configuration page in the following two ways.

• Method 1: On the dataset editing page, click Q&A configuration, as shown in the following figure.

  

• Method 2: On the Create a dataset page, click Q&A configuration, as shown in the following figure.

  

Basic information

On the Q&A configuration page, configure the Basic information.

• You can modify the dataset Display name.

  Note

  You can configure a display name for the dataset that is easy for business users to understand. This name is visible to visitors and helps them understand the data content, such as "2023 Revenue Data by Industry".

• Description

  Provide a simple description to help users find the dataset.

• Dataset type

  Selecting a dataset type helps Q-Copilot understand your data structure, which improves answer accuracy. Supported types include Detail Table, Multi-metric Periodical Table, Key-Value Table, and Other.

  • Detail Table

    Displays detailed data with one record per row. Each record contains multiple dimension values or metric information, such as "Order ID", "User ID", "Order Status", and "Order amount".

  • Multi-metric Periodical Table

    Displays statistical values of metrics over different periods, such as "7-day cumulative sales", "15-day cumulative sales", and "30-day cumulative sales".

  • Key-value table

    A key-value table that includes fields for date, dimension, metric name, and metric value. Examples include "Statistics Date", "KPI Metric Name", "KPI Actual Value", and "KPI Target Value".

• Click Learn now so the system can learn the dataset.

  

  After the learning process is complete, you can click Relearn if the dataset changes.

  

• In Advanced Configuration, set the Dimension Value Matching Mode. Automatic mode and Custom mode are supported.

  

  If a user's question contains a dimension value that the system has not learned because the total number of dimension values exceeds the learning limit, two matching policies are supported:

  • Automatic mode: The system automatically determines whether to rewrite the user's dimension value to a similar, learned dimension value.

  • Custom mode: The administrator can set whether to enable rewriting for each dimension individually.

    • Enable rewriting: Allows the system to map the user's dimension value to a learned dimension value.

    • Disable rewriting: Strictly matches the user's input without rewriting.

• Click Next to go to the Field quality assessment page.

Field quality assessment

On the field quality assessment page, the system evaluates the quality of the fields in the current dataset to improve the Q&A results.

• Click Start assessment.

  Note

  The field quality assessment takes about one to two minutes. You can proceed to the next steps while the assessment is running. You will be notified when the assessment is complete.

• After the field quality assessment is complete, the system provides modification suggestions. You can choose whether to accept them.

  

• Click Apply and relearn to apply the changes to the dataset field information. Then, click Next to go to the Quick questions page.

  

Quick questions

Recommended questions appear after a user selects a dataset to help them get started quickly. Three modes are supported: System recommendations, Expert customization, and Object-based recommendations.

• System recommendations

  You can preview the quick questions. Click Refresh to obtain a new batch of quick questions.

• Expert customization

  In expert customization mode, click Add question to enter the recommended questions that you want to display to users. By default, the first four recommended questions are displayed. If you enter more than four, users can click Refresh to view the others.

  

  To add multiple quick questions, click Batch add and enter them.

  Note

  Enter one question per line. You can add up to 10 questions.

  

• Object-based recommendations

  In Object-based recommendations mode, you can add rules by following these steps.

  1. Click Add recommendation rule or Add rule in the lower-left corner.

     

  2. On the Add recommendation rule page, enter a rule name, target object, and recommended questions.

     1. Recommendation rule name: Enter a name for the rule for easy identification.

     2. Target object: Select the target users.

     3. Recommended questions: Click Add question to add a single question, or click Batch add to add multiple recommended questions.

        Note

        You can add up to 10 questions.

        

  3. For other users, select System recommendations or Expert customization as the applicable rule.

     

Click Confirm changes to complete the configuration.

After you finish, you can go to the Q&A permissions configuration page by clicking Go to Q-Copilot permission management or the Q&A permissions tab. On this page, you can grant permissions to users for this Q&A dataset. For more information, see Q&A permission configuration.

Q&A permission configuration

After you configure the Q&A settings, you can manage Q&A permissions.

1. The following figure shows the Q&A permissions configuration page.

   

2. On the Q&A permissions configuration page, click Add authorization.

   

   You can also click Add authorization on the previous Q&A configuration page.

   

3. Select users to authorize. You can also set an expiration date.

   

4. Click Done. You can now view and manage the authorized users for this Q&A dataset.

   

   If you have the Centralized management permission, you can perform additional permission management tasks on the Q-Copilot > Permission management page. For more information, see Permission Management.

Knowledge base management

The knowledge base is used to configure enterprise-specific knowledge and terminology. After configuration, the model learns this knowledge and uses it for data retrieval and analysis. You can manage the knowledge base on the dataset editing page, where you can configure rules for business logic and regular expression matching.

Access

Go to the Knowledge base management page, as shown in the following figure.

Add business logic

On the Knowledge base management > From current dataset > Business logic tab, you can add business logic.

1. In the upper-right corner, click Add business logic.

2. Enter a Business definition, Data explanation, and Synonyms. Then, configure the Enable forced rewriting option.

   • Business definition: Defines a general concept within the enterprise, such as sales progress or fiscal year. It can be up to 100 characters long. This field must be globally unique. You can enter frequently used Q&A terms here.

   • Data explanation: Provides a specific explanation of the business definition and associates it with data metrics. This helps the model identify and understand different metrics. It can be up to 3000 characters long.

   • Synonyms: Defines different names for the business term within the enterprise. This helps the model recognize different ways of asking questions.

   • Enable forced rewriting: If you enable this option, when a user's question matches the Business definition or its Synonyms, the question is rewritten as the content in Data explanation. Use this option with caution.

3. Click Save.

   To continue adding entries, click Save and add another.

Add a regular expression match

On the Knowledge base management > From current dataset > Regular expression match tab, you can add a regular expression match.

1. In the upper-right corner, click Add regular expression match.

2. Enter a Business definition, Regular expression, and Data explanation. Then, configure the Enable forced rewriting option.

   • Business definition: Used only to identify the name of the regular expression. It is not used for matching user questions. It can be up to 100 characters long and must be unique within the dataset.

   • Regular expression: Used to identify user questions. The operation is executed based on the selected application method. Write the expression in Python style. It can be up to 100 characters long.

     You can enter text to match to test the expression and view the matching result.

   • Data explanation: Provides a specific description of the corresponding content in the regular expression. When combined with the application method, it can explain or rewrite the expression content.

   • Enable forced rewriting: If you enable this option, when a user's question matches the regular expression, the question is rewritten as the content in Data explanation. Use this option with caution.

3. Click Save.

   To continue adding entries, click Save and add another.

Manage the knowledge base

You can manage the knowledge base on the Knowledge base management > From current dataset tab.

1. You can view the status of a knowledge entry in the Enabled column.

   • If the icon for a knowledge entry is , the entry is enabled. You can click the icon to disable it.

   • If the icon for a knowledge entry is , the entry is disabled. You can click the icon to enable it.

2. Click the icon to the right of the target knowledge entry to edit it.

3. Click the icon to the right of the target knowledge entry to delete it.

   You can select multiple entries and delete them in a batch.

4. Click the From enterprise knowledge base tab to view knowledge from the enterprise knowledge base that applies to this dataset.

Fluctuation attribution configuration

On the dataset editing page, click Advanced Configuration and select Fluctuation Attribution to go to the fluctuation attribution configuration page.

On the fluctuation attribution configuration page, you can configure the analysis approach and report template for metrics. For more information, see Key Driver Analysis.

Dataset configuration recommendations

1. For calculable dimension attributes, set the default aggregation method to average. Otherwise, subsequent data calculations will be affected.

   1. Attributes such as price, height, and width involve aggregate calculations, such as maximum, minimum, and average. These attributes should be treated as measures. To prevent automatic summation during queries, setting the default aggregation method to average is more logical. For example: 'What is the sales amount for each brand with car models priced over 300,000?'

2. Specify the data unit in the data field. Otherwise, filtering will be affected.

   1. In the query 'What is the sales amount for brands with prices over 100,000?', if the unit of the price field is 10,000, the backend calculation uses the condition >10 instead of >100,000.

3. Add calculated fields for frequently used dimension counts. The current version does not support counting dimension fields.

   1. The query 'How many customers had sales over 10,000 in each province in 2020?' can be answered correctly if you create a calculated field in the dataset. The system automatically removes duplicates based on the calculation input.

4. Basic principles for configuring dataset field names and descriptions.

   1. Field names:

      1. Field names must be clear, standardized, and align with how users ask questions. Avoid duplicate field names.

      2. Do not use the underlying English field names directly. Avoid unnecessary comments.

      3. Avoid specific time information, such as 'Last 1 day', which can be ambiguous.

   2. Field types: For date/time data, make sure to change the field type to Date in the dataset. Otherwise, the system may not recognize them correctly. Other field types must also be changed to their corresponding types, such as geographic dimensions.

   3. Field aggregation method: For measures, select an appropriate default aggregation method. When a user does not specify an aggregation method, the model uses this configuration. For example, for 'conversion rate', you can select average as the default based on business semantics. For 'cumulative XX', select average or maximum as the default, not sum.