Semantic View
A Semantic View is the business semantic layer in the Lakehouse, creating an abstraction layer between physical table structures and business analysis requirements. It solves the problem of "the same metric having different definitions across different reports."
What Problem Does It Solve
Without a semantic view, different reports and applications write their own JOIN and aggregation logic, and the same "monthly active users" might be calculated differently in different places. A semantic view centrally manages metric definitions, ensuring a unified definition across the entire organization.
Comparison with Regular Views
| Comparison Item | Regular View | Semantic View |
|---|---|---|
| Definition method | SQL query | Declarative (table relationships + dimensions + metrics) |
| Primary use | Logic encapsulation, permission isolation | Business semantic layer, AI data access |
| JOIN handling | Manual | Engine handles automatically |
| Target audience | Developers | Business users, AI Agents |
Core Components
- Logical Tables (TABLES): Map to physical tables, declare primary/foreign key relationships, and automatically JOIN during queries
- Dimensions (DIMENSIONS): Categorical attributes such as region, time, and product category
- Metrics (METRICS): Aggregated measures such as sales amount, user count, and conversion rate
- Filters (FILTERS): Reusable predefined filter conditions
Typical Scenarios
| Scenario | Description |
|---|---|
| Unified metric definitions | Centrally define KPIs to avoid inconsistencies in calculation logic across reports |
| Lower query barrier | Business users query using business terms without needing to understand physical table structures |
| AI Agent data access | Serve as the semantic data source for Analytics Agent, enabling natural language Q&A |