The intelligent analysis agent (hereinafter referred to as the analysis agent or intelligent agent) is an expert advisor based on AI deep reasoning and big data analysis. The analysis agent includes the data query agent and the deep research agent. The data query agent, as a flexibly configurable intelligent data query tool, enables users to accurately query business data through simple conversations. It supports querying across multiple datasets, semantic model parsing, business knowledge invocation, question recommendation and bookmarking, multi-turn interactive data querying, as well as attribution analysis, fully meeting users' diverse data query needs.

The data query agent feature of the intelligent analysis agent can accurately interpret user intent through natural language conversations, enabling collaborative queries across multiple datasets and in-depth attribution analysis. Supplemented by personalized recommendations and multi-turn interactions, it provides users with efficient, accurate, and unified data query services.

Specific features #

• Recommended questions: The intelligent agent is preloaded with questions that most team members care about, allowing other users to quickly query by clicking on the questions.
• Bookmarked questions: Users can bookmark frequently asked questions for quick access in future queries by simply clicking.
• Data query: Freely query data within the intelligent agent through conversations, supporting multi-turn interactions.
• Attribution analysis: The web-based intelligent agent integrates attribution analysis capabilities, enabling quick analysis of data while querying.
• Scheduled push: Questions are pushed to relevant users at a frequency set by the user.

Application scenarios #

• Business employees in enterprises who currently lack data analysis capabilities and statistical knowledge can use the intelligent agent to lower the threshold for querying and understanding data.
• Managers and business personnel in enterprises may not be familiar with datasets or the underlying data processing logic. When performing data queries and business analysis, they want to quickly access the data they need. The intelligent agent enables users to obtain answers simply by asking questions, without needing to select datasets.
• Data/technical teams in enterprises can set the data that business users or managers focus on daily as recommended questions for the intelligent agent. This way, users can quickly view data in Feishu (IM platform) without opening the product.

• Project creation and user authorization have been completed. For details, see Project Information, User Management
• Data preparation has been completed. For details, see Data Preparation Overview.
• Intelligent agent configuration has been completed. For details, seeOverview of the Creation and Configuration Process

Currently, the Data Query Agent supports the following types of datasets:

• Supports extracted datasets
• Supports direct connection to ClickHouse/ByteHouse datasets
• Supports direct connection to Doris/selectdb/starrocks datasets
• Supports direct connection to Postgres/hologres/openguass
• Supports direct connection to mysql

1. After logging in to the Volcano Engine DataAgent-Analysis Agent page with your account and password, select the published intelligent agent on the left side of the page to enter the conversation page and experience the data query feature.

2. After entering the intelligent agent conversation interface, click the switch button in the dialog box to switch between the "Smart Query" and "In-depth Research" features. The Data Query Agent only provides the "Smart Query" feature.

3. Users can click the dialog box to select built-in questions to ask, or directly enter questions to ask.

   Natural language data query	Quickly view recommended questions
   The main usage of the agent allows users to directly describe their data query requirements in text to the agent. After receiving a response, users can click "Follow-up" to continue interacting with the agent regarding related questions. When querying data with the agent, there is no need to select a dataset. The agent will automatically query data from the configured datasets.	Recommended questions are provided below the dialog box. Users can click any recommended question to directly query data. Recommended questions are automatically generated based on users' historical queries and can also be customized. For specific configuration methods, see Overview of the Creation and Configuration Process.

Common chart analysis #

You can choose visualization chart display methods, including metric cards, line charts, bar charts, pie charts, combination charts, tables, and more.

Data interpretation #

1. Click the "Data Interpretation" button in the upper right corner of the returned results. The Intelligent Analysis Agent will interpret the data based on the query results.
2. After the interpretation is complete, the interpretation results will be returned to you. You can also click "View Details" to see the detailed data interpretation.

Anomaly detection #

1. Click the "Anomaly detection" button in the upper right corner of the returned results. The intelligent analysis Agent will perform anomaly detection on the queried results.

2. After the detection is complete, the detection results will be returned to you.

Attribution analysis #

When abnormal data is found in conventional chart analysis, you can use attribution analysis to investigate the abnormal results. The Agent can automatically conduct time series analysis on the influencing factors of abnormal metrics, displaying the top three positive contributors and the top three negative contributors, along with their corresponding contribution values, contribution rates, comparison differences, and comparison percentages, enabling intelligent attribution analysis.

1. Click the "Attribution analysis" button in the upper right corner of the chart for the query results. In the pop-up window, the intelligent analysis Agent pre-fills the attribution metrics, dimensions, and baseline/observation dates for you. You can also manually adjust the relevant attribution parameters.
   

   Parameters	Note
   Attribution metric & attribution dimension	The intelligent agent will pre-fill an attribution metric and attribution dimension based on the chart, and you can reselect them in the dropdown menu.
   Reference date	Click to set the attribution "reference date". The reference date is a specific date used as a standard for comparison, to measure and compare data changes relative to the observation date. The reference date must be earlier than the observation date.
   Observation date	Click to set the attribution "observation date". The observation date is the specific date for data collection and phenomenon observation. The observation date must not be earlier than the reference date.

2. After the attribution analysis is complete, the Agent will provide the attribution results and display the top three positive contributors and the top three negative contributors, along with their corresponding contribution values, contribution rates, comparison differences, and comparison percentages.

• Top 3 positive contributors: Influencing factors that have a positive impact on metric changes and are ranked among the top three contributors.
• Top 3 negative contributors: Influencing factors that have a negative impact on metric changes and are ranked among the top three contributors.
• For the algorithms of contribution value and contribution rate, refer to: Dimension attribution - algorithm.

View analysis process #

You can choose whether to view the reasoning process, analysis parameters, and other content.

AI follow-up questions #

Reset conversation #

While using the Agent, the Agent will automatically use historical conversation data as input for intelligent queries. If you do not need to refer to historical conversation records during interaction with the Agent, you can click "Reset conversation" to ask again. Subsequent new conversations will no longer be associated with the above historical conversation content.

View conversation history & start a new conversation #

• Allow users to view and delete historical Q&A in recent conversations.
  
• Allow users to create new conversation windows.
  

Download & save questions #

The assistant allows users to download and save previously asked questions.

• Download: Currently, query results can be downloaded in two formats: xlsx and Feishu Sheets.

  tip

  When downloading as a Feishu spreadsheet, you must complete the Feishu integration configuration first.

• Save: Saved questions will be displayed on the assistant's conversation homepage, so users do not need to manually enter the same question again in the future.

Scheduled push #

The Feishu assistant supports proactive push settings, which means users can configure routine pushes for the assistant, such as automatically pushing desired data to users at a scheduled time each day.

When data query results are inaccurate, unstable, or have other quality issues, optimization can be considered from three aspects: the quality of the original dataset, the assistant's configuration, and the quality of the question description.
When optimizing the assistant, the recommended order of priority is: dataset quality (dataset governance) > dataset-semantic model definition > dataset-knowledge base configuration > question accuracy. That is, first prioritize optimizing the quality of datasets involved in business scenarios, then optimize the assistant's knowledge configuration, and finally optimize based on the question's configuration or description.
Disclaimer
(1) All content output by this service is generated by artificial intelligence models.
(2) Intended users of this feature: Only for customers' internal, proprietary business, and personal use. Customers must not provide generative AI services to the public through the data assistant.
(3) This product supports both SaaS and on-premises deployment versions. For purchase or trial, please contact sales personnel for consultation.