Return to news

How to make natural language data analysis credible

The semantic layer, indicator caliber, query permissions and result verification are necessary conditions for natural language analysis to enter enterprise scenarios.

product perspective6 minutes

Models need to understand business semantics

Database field names do not equate to business language. Natural language analysis requires establishing mappings between metrics, dimensions, business terms, and data tables.

The indicator caliber must be unified

There may be different calculation methods for indicators such as sales, effective customers, and conversion rates. The system should prioritize the use of confirmed indicator definitions rather than ad hoc generated calibers.

Query scope is controlled by identity and data permissions

Users can only query the data range they have access to, and permission checks and sensitive field controls need to be performed before and after generating queries.

Allow results to be verified

The results should also display the query conditions, time range, indicator caliber and necessary data source description.

For complex analysis, query records and analysis paths can be retained to facilitate review and continuous improvement.