Auto-keyword Auto-question
Use a chat model to generate keywords and questions from the original chunks.
When selecting a chunking method, you can also enable auto-keyword or auto-question generation to increase retrieval rates. This feature uses a chat model to produce a specified number of keywords and questions from each created chunk, creating a layer of higher-level information from the original content.
Enabling this feature increases document indexing time, as all created chunks will be sent to the chat model for keyword or question generation.
-
Auto-keyword
- Definition: The number of additional keywords the LLM generates for each chunk. By supplying synonyms for text that is unfriendly to tokenization or multilingual content, this improves recall for full-text or hybrid retrieval. It can also be used to correct bad cases. Disabling this can significantly accelerate parsing.
- Common Values:
0
: Disabled;3
-5
= Recommended (if a chunk has over a thousand characters, more keywords may be needed);- Maximum
30
. Note that, as the number increases, the marginal benefit decreases.
-
Auto-question
- Definition: Generates potential FAQ-style questions for each chunk, making retrieval matches more aligned with real user queries (Who/What/Why).
- Common Values:
0
= disabled;1–2
= commonly used (if a chunk has thousands of characters, more may be needed);- Upper limit
30
(to avoid generating too many at once). Can also be used to correct bad cases.
- Typical Use Cases: Scenarios requiring FAQ retrieval, such as product manuals, policy documents, etc.
Configuration
On the Configuration page of your knowledge base, you will find the Auto-keyword and Auto-question sliders under Page rank.
The Auto-keyword or Auto-question value must be an integer. If you set their value to a non-integer, say 1.7, it will be rounded down to the nearest integer, which in this case is 1.
Best practices
If you are uncertain how to set auto-keyword or auto-question values, here are some best practices gathered from our community:
Use cases or typical scenarios | Document volume/length | Auto_keyword (0–30) | Auto_question (0–30) |
---|---|---|---|
1. Internal Process Guidance for Employee Handbook | Small, under 10 pages | 0 | 0 |
2. Customer Service FAQ Hot Questions | Medium, 10–100 pages | 3–7 | 1–3 |
3. Technical Whitepapers: Development Standards, Protocol Explanations | Large, over 100 pages | 2–4 | 1–2 |
4. Contracts / Regulations / Legal Clause Retrieval | Large, over 50 pages | 2–5 | 0–1 |
5. Multi-repository Layered New Documents + Old Archive | Many | Adjust as appropriate | Adjust as appropriate |
6. Social Media Comment Pool: Multilingual & Mixed Spelling | Very large volume of short text | 8–12 | 0 |
7. Operational Logs for DevOps Troubleshooting | Very large volume of short text | 3–6 | 0 |
8. Marketing Asset Library: Multilingual Product Descriptions | Medium | 6–10 | 1–2 |
9. Training Courseware / eBooks | Large | 2–5 | 1–2 |
10. Maintenance Manual: Equipment Diagrams + Steps | Medium | 3–7 | 1–2 |