Data Format for This Category
Daily documents for channel attribution and lead distribution scenarios include attribution rules, lead distribution tables, channel reports, and exception checklists. Core fields include channel identifier, lead ID, assigned sales ID, conversion amount, parameter validation flag, and others. Common units are unitless, yuan, and count. Data sources cover ad platforms, CRM systems, attribution SDKs, and form systems. Most data updates occur daily or weekly. The number of entries per document fluctuates with business scale. Rapid retrieval of the latest rules and report content is required to handle exceptions and inquiries in operational on-call and internal ticket scenarios.
Constraints Imposed on Knowledge Base Retrieval and Recall
Attribution rules require precise matching of channel parameter combinations. Retrieved results must strictly align with field relationships. Deviations in matching can distort channel ROI calculations and impact internal budget allocation. Lead distribution table entries increase as businesses grow. The number of results retrieved in a single call must be controlled within a reasonable range. Excessive results can cause context overflow and disrupt LLM output, delaying operational ticket processing. Daily updated channel reports must support rapid retrieval of the past 7 days of data. The system must adapt to the synchronization rhythm of frequently updated data sources. This ensures operational personnel retrieve the latest data to complete exception checks. Field validation logic for exception checklists must fully match the rule base in the knowledge base. Retrieved results must include the latest validation rule versions. Using outdated rules to process tickets can lead to disputes over sales assignments.
Configuration Settings
This page is written based on FastGPT v4.17.0.
| Configuration Item | Recommended Value | Rationale |
|---|---|---|
similarity | 0.75–0.85 | Channel parameter verification requires precise matching of identifier combinations. A threshold that is too low will introduce irrelevant rules, while a threshold that is too high may miss valid matching items. This value is calibrated based on actual business testing. |
limit | Top 6 entries | Lead distribution tables have a large number of entries per batch. Excessive results will exceed the context window, while too few results will fail to cover all associated leads. This setting adapts to the needs of rapid troubleshooting during operational on-call shifts. |
searchMode | Vector search + keyword hybrid | Attribution rules include both structured parameters and natural language descriptions. The hybrid mode balances precise matching and semantic understanding, adapting to retrieval needs for multiple types of documents. |
usingReRank | Enabled | Channel ownership determination requires sorting to match the relevance of sales IDs and lead sources. Reranking improves the display priority of highly relevant results and reduces the time spent on manual verification. |
datasetSearchUsingExtensionQuery | Enabled | Descriptions of operational ticket issues often include vague exception prompts. Extended queries can convert questions into retrieval terms that better fit attribution rules, improving recall accuracy. |
collectionIds | Specify attribution rule base and lead distribution table collection, support dynamic adjustment using workflow variables | Limiting the retrieval scope excludes irrelevant documents, avoids retrieving unrelated content during operational on-call shifts, and improves response efficiency. Variable reference adapts to retrieval needs for different ticket scenarios. |
Three Common Configuration Mistakes
- Phenomenon: The AI reply consistently includes the phrase "FastGPT is a knowledge base question answering system based on large language models (LLM)" at the end. Cause: The default endnote configuration after knowledge base retrieval is not disabled, or the fixed output content from the system is not removed from the prompt template.
- Phenomenon: After deploying via docker-compose, knowledge base and workflow configurations are lost, but interfaces function normally. Cause: Persistent storage volumes are not mounted, and locally stored configuration data is not retained after container restart.
- Phenomenon: The number of retrieval results does not match the configured
limitparameter, with missing or redundant results. Cause: The result counting logic after reranking is not updated synchronously, or threshold parameters in hybrid search mode are not linked withlimit, causing the actual number of retrieved results to deviate from the configured value.
How to Verify Successful Configuration
- Input a known channel parameter verification rule, check whether the retrieved results include the complete content of the rule, and verify field matching accuracy.
- View the number of retrieval results, confirm that it matches the configured
limitvalue, and verify that the quantity control logic is effective. - Submit an operational ticket containing a vague exception description, check whether the extended query is effective, and retrieve relevant attribution rules and exception checklists.
- Attempt to dynamically specify the retrieval knowledge base collection using workflow variables, verify that the variable reference function is effective, and that retrieval results meet expectations.
- Check the AI reply content, confirm that the fixed system endnote phrase is not included, and verify the collaborative effect of the prompt and retrieval configuration.
The values provided are common starting points and should be measured against the reader's own samples.
Question material comes from public community discussions. Configuration values are common starting points and should be measured against your own samples. Verified on 2026-10-19.