Table question answering
Concepts
Table question answering (Table QA) provides AI chat capabilities for questions that require queries and inference based on a given table. Technically, Table QA uses natural language processing (NLP) to transform a user's question into an SQL statement. This statement is then executed to query and infer information from the table data. The system uses the results to answer the user's question. The process is shown in the following figure:
Currently, only the Enterprise Edition and Ultimate Edition support the Table QA feature.
Platform as a Service (PaaS) API operations for Table QA are not yet available.
Scenarios
The Table QA feature is ideal for scenarios where you have many business tables and user questions can be answered by retrieving information from these tables. An example is shown in the following figure:
In these scenarios, you can simply import the tables to enable automatic responses. You can further improve performance by optimizing the Q&A settings.
Compared to other engines such as FAQ and conversation flows, Table QA reduces knowledge maintenance costs and provides stronger inference and calculation capabilities.
Features
Table QA supports the following capabilities:
Question type | Supported question format | Example question |
Simple query | Single-property query | What is the process size of the YiTian 710 processor? |
Multi-property query | What is the minimum investment for "Wenjian No. 12"? What is its rate of return? | |
Constrained query | Property-constrained query | Which wealth management products have a rate of return greater than 5%? |
Property-sorted query | What are the three largest reservoirs in China by capacity? | |
Inference and calculation | Aggregation (count, average, etc.) | What is the average rate of return for principal-protected wealth management products? |
Conditional calculations | Does the YiTian 710 processor support DDR5 memory? | |
Comparative calculation | How much more expensive is the iPhone 14 Pro than the iPhone 13 Pro? | |
Grouped statistics | Which three provinces have the most small-sized reservoirs? | |
Column calculation | What is the total score of Zhang Dazhu from Class 1, Grade 3, across all subjects in the final exam? | |
Multi-turn conversation | Information collection through bot follow-up questions | How much is the E63? / The elegant model. |
Multi-turn conversation about a table | Wealth management products I can buy with 50,000 yuan. / The low-risk ones. / What if risk is not a concern? | |
Open capabilities | Integrate with environment variables | Supports using environment variables to influence table selection, table header location, and more. |
Provide PaaS API database operations | Provides API operations for real-time batch creation, deletion, and modification of data. | |
Cross-table query | Multi-table join query | Supports queries that join multiple tables. |
Feature usage workflow
Before using Table QA, create a bot instance. Using Table QA involves the following steps:
For more information about importing table data, see Table data management.
For more information about configuring Q&A policies, see Table QA intervention.