After knowledge building is complete, the large language model organizes knowledge content into a Wiki and a knowledge graph. This topic describes how to browse the Wiki directory, explore the knowledge graph and graph reasoning, use unified knowledge search, and use a knowledge base in Data Agent and AI assistant services. For information about how to create and build a knowledge base, see Create and build a knowledge base.
Feature overview
After knowledge building is complete, you can use the knowledge in a knowledge base in the following ways:
Wiki: Organizes knowledge content into directories, topics, and pages for easy browsing and navigation.
Graph: Displays knowledge objects and their relationships as an explorable network.
Graph reasoning: Finds relevant paths between knowledge objects based on business questions and provides supporting evidence.
Unified knowledge search: Searches knowledge objects, Wiki pages, and graph entities from a single entry point.
Intelligent Q&A: Enters Data Agent intelligent Q&A with the knowledge base context, or grants access to the knowledge base for use by AI assistant services and Data Agent.
Knowledge bases are centrally created and managed by an independent knowledge base module and can be used in AI assistant services and Data Agent. For information about how to create a knowledge base, configure sources, build knowledge, and manage permissions, see Create and build a knowledge base.
Prerequisites
Knowledge base status: At least one successful knowledge building task has been completed. If a legacy knowledge base has not been upgraded, you must first complete the knowledge base upgrade.
Access permissions:
Enterprise knowledge base: All tenant members have view and search permissions by default.
Personal knowledge base: The knowledge base owner or administrator must grant access through sharing.
Key concept
Concept | Description |
Wiki | A knowledge collection generated from the knowledge base content, organized by business domain, topic, and page hierarchy. |
Knowledge graph | A network that visualizes knowledge objects and their relationships, showing dependencies, containment, references, and semantic relationships. |
Graph reasoning | Searches for association paths between knowledge objects based on specific business questions and outputs supporting evidence. |
Knowledge search | A search entry point that enables centralized retrieval of knowledge objects, Wiki pages, and graph entities within the current knowledge base. |
Knowledge base overview
Go to the overview of the knowledge base details page to view the data asset accumulation and building status of the current knowledge base:
Statistics: Shows the count of various knowledge objects such as tables, entities, metrics, and dimensions.
Activity dashboard: Shows recently generated topic knowledge, knowledge objects, and recent building records.
If the statistics are empty or a message indicates that data is not ready, this means that no valid building has been completed. You must first initiate a knowledge building task. For more information, see Create and build a knowledge base.
Browse and search the Wiki
On the knowledge base details page, go to Wiki > Directory to browse knowledge by business domain hierarchy.
Browse and search
Expand or collapse nodes in the directory to quickly locate business topics.
Use Wiki search to find content by page title, directory, section, or body keywords.
Search results display the matched Wiki pages and matched sections. You can navigate to a page to view its full content.
On the page details, you can view the summary, content type, language, number of sources, topic tags, and reference information.
View sources and references
Wiki pages display sources and evidence. Sources can come from semantic models, platform assets, or uploaded documents.
Click a reference to view the corresponding evidence and source information.
For sources that support preview, you can view the original text or navigate to the original location.
Some sources only support traceability and do not support online preview. This does not affect the use of the corresponding Wiki content.
If the knowledge index is still being synchronized, refresh or retry later.
Use knowledge graphs and graph reasoning
On the knowledge base details page, go to the Graph tab of the Wiki. Enter an object name, business question, or natural language description, and select a query scenario to locate knowledge paths around specific business questions.
Search the graph
In the search box, enter an object name, business question, or natural language description, and then select a query scenario. Common scenarios are as follows:
Query scenario | Example question |
Metrics and table columns | Which tables are used to calculate GMV? |
Production lineage | Where is this table produced from? |
Knowledge evidence | Where does this piece of knowledge come from? |
Impact analysis | What will be affected if this table is modified? |
Metric explanation | What is the definition and calculation basis of GMV? |
Join discovery | Which tables can the order table be joined with? |
Governance check | What quality or security requirements does this table have? |
Business domain scope | What topics and data assets are included in the transaction domain? |
After you perform a search, the page displays candidate knowledge objects, nodes, and relationships. If multiple candidate objects are returned, select the candidate that best matches your question, and then view the graph results.
Read the graph
Nodes represent knowledge objects, such as workspaces, projects, tables, columns, metrics, data domains, documents, or knowledge evidence.
Edges represent relationships between objects, such as "contains", "depends on", "queries table", "contains metric", "belongs to data domain", and "serves as evidence".
After you select a node, you can view the node details, relationships, and source evidence.
Click Expand or Continue Expanding to load more related objects. If too many results are returned, the page may display only some of the relationships.
You can use filter, zoom, reset layout, and full-screen operations to organize the view.
View reasoning paths
Graph reasoning breaks down a question into multiple association paths to help you understand the derivation process of a conclusion. When reading a reasoning path, we recommend that you review the following in order:
Starting object: The metric, table, column, or business topic that the question involves.
Intermediate relationships: The dependency, containment, reference, or semantic relationships between objects.
Target object: The knowledge, asset, or conclusion that is ultimately associated.
Evidence source: The wiki, semantic object, platform asset, or document that supports the relationship or conclusion.
Unified knowledge search
At the top of any page in the knowledge base, you can expand the knowledge search bar to perform a global search across the knowledge base. Enter a question or keyword and submit it to view search results in the current knowledge base. If there are many results, you can navigate through pages. Knowledge search supports business terms, technical names, file names, or natural language questions, and simultaneously searches for the following content in the current knowledge base:
Knowledge objects and their complete details.
Wiki pages and matched sections.
Graph entities, relationships, and inference results.
Search examples:
What does transaction amount meanTables related to user profilesColumns containing phone numbersData quality inspection processDaily aggregation query example for the orders table
The search scope is limited to the current knowledge base and does not automatically query across knowledge bases. When using search results, note the following:
Evaluate whether results are applicable based on the knowledge base, business domain, object type, and source.
Click a result to view the knowledge object details or original materials and verify important conclusions.
The system does not modify or translate physical table names, column names, SQL statements, code, or IDs.
You can only search knowledge bases that you have access to.
Use knowledge bases in intelligent applications
On the knowledge base details page, click Knowledge Assistant to start an intelligent Q&A session with the current knowledge base as context. You can use natural language to ask about business definitions, data assets, and document content.
You can also use knowledge bases in Data Agent and AI Assistant Service. The available knowledge bases depend on the knowledge base type and authorization status:
Enterprise knowledge base: Available to all members under the same tenant.
Personal knowledge base: Available only to the owner by default. To make it available to Data Agent or AI Assistant Service, the owner or administrator must grant access to the corresponding instance through Share. For more information, see Permission management.
Data Agent and AI Assistant Service cannot serve as knowledge base administrators. They can only use a knowledge base after being granted access through Share.
FAQ
Q: Why do some pages have no data?
A: The system generates knowledge only when source data exists and is successfully analyzed. For example, no quality rules are displayed if none exist, and cross-table query examples may not be generated if no reliable relationships are found. An empty page does not necessarily indicate a build failure.
Q: Why is the knowledge still in Chinese after I switch to the English interface?
A: The language of knowledge content is determined during the initial build. Switching the console language does not translate or change existing knowledge. For more information, see Language settings.
Q: Why can't I find the expected results in search?
A: The search scope is limited to the current knowledge base, and you can only search knowledge bases that you have access to. Verify that the assets or documents corresponding to your keywords have been successfully built and deployed. Uploaded files can only be searched after they are successfully built. For more information, see Build a knowledge base.
Q: Why was the coverage not shown in green even though it already displays 100.0%?
A: The interface displays coverage rounded to one decimal place. As long as the rounded value shows
100.0%, the value and progress bar are displayed in the completed (green) state.Q: Why does the graph show only some relationships?
A: The graph page limits the number of nodes and relationships expanded at a time to ensure page responsiveness. You can use Expand More, apply filter conditions, or switch query scenarios to explore further.
Q: Why doesn't a newly added document appear in search results?
A: Adding a source only completes the source configuration. You still need to create and successfully run the corresponding knowledge build task. After the build is complete, you need to wait for the knowledge index to synchronize before the content becomes searchable.
Best practices
Evaluate whether search results are applicable based on the knowledge base, business domain, object type, and source. For important conclusions, click to view the knowledge object details or original materials for verification.
Names, descriptions, and relationships inferred by the system are for reference only. Important business definitions should be confirmed by the business owner.
Regularly check request counts and token consumption in Usage Management on the knowledge base list page to evaluate knowledge base usage and costs.
Use a knowledge base within the scope of the granted access. Users who are granted access can only use the knowledge base for search or Q&A, but cannot modify the knowledge base.
Select a graph query scenario based on the specific business question. First confirm the candidate objects, then review the reasoning chain, and gradually expand the graph to control its scale.