You need to enable JavaScript to run this app.
DataAgent (OnPremise)

DataAgent (OnPremise)

  • Documentation
    • DataAgent (OnPremise)Intelligent Analysis Agent User GuideBest PracticesConfiguration and Tuning PracticesBest Practices: Recommendations and Examples for Intelligent Data Query Questions
Copy page
Download PDF
Configuration and Tuning Practices
Best Practices: Recommendations and Examples for Intelligent Data Query Questions
Copy page
Download PDF
Best Practices: Recommendations and Examples for Intelligent Data Query Questions

Question guidelines and examples

A good question should include the following key elements:

Image

Elements

Explanation

Time

Specify the time point or range for the data, such as "the past 7 days", "Labor Day last year", "within the past 30 days", or "the week of January 18".

Metrics

Specify one or more metrics to query, such as "GMV" or "order volume".

Filter conditions (optional)

Specify the statistical dimension for the data, such as "by province" or "by primary and secondary category".

Statistical method (optional)

Specify the statistical method for the data, such as "How many XX are there?", "sum", "average", "maximum", "week-over-week", "year-over-year", and more.

Presentation method (optional)

Supports tables as well as common data visualizations such as line charts, bar charts, pie charts, and more.

Question description and recommendations

Scenario 1: Querying or processing data in a dataset

Case 1: Querying specific data

  • Description of intelligent data query capabilities: When querying data from a specific table field in a dataset, the question does not need to exactly match the field name in the table. The large model will match the appropriate field based on its understanding.
  • Configuration/question recommendations: When multiple field names in the dataset are very similar, it is recommended to:
    • In the agent's dataset configuration: You can use the semantic configuration capabilities of the dataset to further configure synonyms and characteristics for each field, helping the large model understand the differences between fields and improve the accuracy of field matching.
    • When asking a data query question: The more detailed the description of the field you want to query in the question, the more accurate the field matching will be.

Case 2: Performing calculations on dataset fields

Detailed scenarios

Question recommendations

Calculations for maximum, minimum, sorting, period-over-period comparisons, and more.

For period-over-period comparisons, it is best to specify the time granularity for comparison.

Understanding formulas for common calculated metrics (such as xxx rate)

The general knowledge of large models can interpret some calculated metrics in certain scenarios. For example, a large model can understand "profit margin" = total profit / total sales; however, other xxx rates may not be within the general knowledge scope of the large model, which may result in misinterpretation.
Therefore, when dealing with very common calculated metrics, you can directly ask questions. If you find that the large model misunderstands the metric in the answer, it is recommended to process the metric calculation directly based on the dataset and prepare the metric in advance.

Scenario 2: Ambiguous questions

Detailed scenarios

Question recommendations

Very vague questions

Avoid very vague questions.
For example: Take a look at the situation in each region. The large model cannot accurately match which metric you need to analyze.

Questions with unclear intent

If you find many questions with unclear intent when using the intelligent analysis Agent, you can try the following:

  • After intent clarification is configured, the intelligent agent will proactively ask follow-up questions about ambiguous intent in subsequent queries to clarify the user's question.
  • If ambiguous intent questions are common and typical, you can configure ambiguous questions to associate them with a clear question, improving the large model's understanding of similar ambiguous intent questions.

Cannot recall specific dimension values

When the dimension values that need to be filtered are unclear, you can use "contains" to ask questions.
For example, "Author nickname: Xiaoming, number of followers on the day greater than 100,000, what is the xxx tag?" ---> "Author nicknamecontains Xiaoming, number of followers on the day greater than 100,000, what is the xxx tag?".

Scenario 3: Personnel references

  • The intelligent analysis Agent currently cannot proactively identify personnel identities, so you should avoid questions such as “What is the sales revenue for the region I am responsible for” that contain unclear personnel references (such as “I”).
  • It is recommended to directly specify the person's name in the question description, for example: “What is the sales revenue for the region Liu You is responsible for” or “What is the sales revenue for the region managed by regional manager Liu You”.

Scenario 4: Multiple intents

When asking questions using the intelligent analysis Agent, do not include multiple intents in a single interaction. That is, each question should query only one piece of data.
Incorrect example: Which are the top 10 products with the highest profit in 2024? What are the year-over-year and month-over-month changes in profit and sales revenue for these products? Which product has the largest year-over-year increase? ——In this example, a single question interaction contains multiple detailed intents.

Last updated: 2026.06.02 14:59:27