Exploring Query Generation
Summarize
Summary of Exploring Query Generation
Query Generation is an AI-powered ServiceNow feature that translates user questions into executable queries, which include data source selection, filtering, aggregation, and visualization instructions. The output includes a textual summary, data visualization, and suggestions for further exploration, enabling users to quickly gain insights from data.
Show less
This capability relies on a semantic data layer, a streamlined representation of tables (entities) and columns (dimensions), to accurately map user questions to the relevant data without overwhelming system resources or generating inaccurate results.
How Query Generation Works
Before invoking the large language model (LLM), Query Generation filters the instance schema to focus only on the most relevant entities and dimensions related to the user’s question. This approach:
- Prevents hallucination by the LLM by providing precise, accurate metadata about available tables and columns.
- Improves performance and accuracy by limiting the context window to the top 2 matching entities and top 30 matching dimensions for the query.
The filtered schema is passed to the LLM, which generates a semantic query. This semantic query is then converted into an executable query that can be run against the data.
Configuration and Management
Not all fact tables are included by default to avoid performance degradation. Administrators can view and modify which tables are enabled for semantic generation via the Semantic Tables Configurations list, but should do so cautiously due to potential performance impacts.
Users and Roles
- ServiceNow AI Platform administrators: Manage the semantic data layer, including adding or removing tables. Only users with admin roles can modify Query Generation configurations.
- End users of Now Assist in Platform Analytics: Access Query Generation indirectly through applications that call the service. They require appropriate roles to utilize these features but do not interact with Query Generation directly.
Next Steps
For practical implementation, customers should explore detailed configuration guidance and reference materials to tailor Query Generation to their environment and use cases.
Query Generation is an AI-powered service that translates user questions into an executable query and returns the results. An executable query contains the data source, filter, aggregation, and visualization instructions that best answer the user's question. The results include a textual summary, a data visualization, and suggestions for follow-up.
Query Generation overview
Query Generation relies on a semantic data layer to generate queries. The semantic data layer is a flat representation of tables and table columns that the Query Generator uses to find the actual facts tables and columns related to a user utterance. Specifically, facts tables are represented by Entity records and their columns by Dimension records.
Not all facts tables are included in Query Generation, as this would overload an instance. To see which facts tables are included, open the Semantic Tables Configurations list [sn_query_gen_table_config_list], and note which tables are present and have Enable Semantic Generation = true. You can add more tables to the list, but be careful of possible performance impacts. For more information, see Add a table to the semantic data layer.
How Query Generation works
- It provides the LLM with precise grounds for truth about available tables and columns, preventing hallucination of non-existent data structures.
- It maintains a focused context window, which improves LLM performance and accuracy compared to processing the entire schema.
Query Generation uses a semantic filter to narrow the entities (facts tables) to the 2 closest matches to the user's question. Then from those entities, it narrows the dimensions (columns) to the 30 most similar to the user's question. Query Generation passes these results to the LLM, which generates a semantic query. A constitutor takes this semantic query and translates it into an executable query.
Query Generation users
| User | Description |
|---|---|
| ServiceNow AI Platform administrators responsible for Now Assist in Platform Analytics [admin] | Administrators can add or remove tables from the semantic data layer. Only users with the admin role can read or change Query Generation records. |
| Users of Now Assist in Platform Analytics applications | Users of the Now Assist in Platform Analytics applications call Query Generation through those applications, although Query Generation is not visible to them. They should have the required Query Generation user roles through the roles granted to them to use the intermediary application. |