What data looks like for this category
Business handling process data for power generation groups is primarily sourced from internal business manuals, on-site execution documents, monthly updated group policy documents, and quarterly adjusted process specifications. Most documents are in docx or pdf format, split by business domain, and include fields such as process number, responsible department, handling time limit (unit: working days), approval nodes, and required material list. The length of individual documents varies widely, ranging from dozens of pages for complete group-level processes to a few pages for on-site specific handling rules. Update cadence follows monthly policy updates and quarterly process adjustments, so incremental sync per update cycle is required.
What constraints these characteristics impose on the knowledge base retrieval and recall link
Business process data for power generation groups has a fixed structure. It includes clear responsible department and time limit fields. Retrieval results must accurately match business nodes to avoid retrieving irrelevant generic content. The wide range of document lengths requires adaptive chunk parsing and multi-granularity recall for long documents, to prevent core steps from being truncated. Fixed, staged update frequency requires regular incremental sync to maintain data timeliness. Process documents across different business domains are independent, so retrieval scope must be isolated by domain to avoid cross-domain interference. Additionally, internal specialized terms exist, so term expansion support is needed to align user questions with internal document consistency.
Configuration Settings
Reference only; verify against the target FastGPT release and deployment documentation.
| Configuration Item | Recommended Setting | Rationale |
|---|---|---|
similarity | 0.75–0.85 | Meets the accuracy requirements for matching business handling processes, avoids retrieving non-process documents |
limit | Top 8 entries | Covers complete steps and associated materials for business handling, avoids redundant or missing results |
searchMode | Full-text retrieval | Adapts to process step splitting in long documents, accurately targets specific handling nodes |
usingReRank | Enabled | Sorts retrieved results by relevance, corrects bias from single-vector retrieval, highlights core process steps |
datasetSearchUsingExtensionQuery | Enabled | Automatically expands synonyms for user questions to match internal specialized terms of the group, improving retrieval accuracy |
collectionIds | Bound by business domain | Binds dedicated knowledge base collections for each business domain of the power generation group, filters retrieval scope based on user permissions |
Three Common Configuration Mistakes
- Phenomenon: Retrieval results include process documents from unauthorized business domains, and the department associated with the displayed process in the interface does not match the current user's permissions. Cause: The
collectionIdsparameter is not correctly bound, and the knowledge base collection is not filtered based on the user's department. - Phenomenon: When batch uploading business process documents in docx or pdf format, some files lose step information after parsing and chunking. Cause: Reasonable retrieval matching logic is not set for long documents, leading to core steps not being fully retrieved.
- Phenomenon: When a user asks "how to handle the new electricity meter installation process", the group's internal "application for installation" related process documents are not retrieved. Cause:
datasetSearchUsingExtensionQueryis not enabled, and no term expansion is performed for user questions, so internal specialized terms of the group cannot be matched.
How to Confirm the Configuration is Correct
- Upload a power generation group business process document, retrieve the corresponding keyword, and check whether the number of retrieved results matches the value set for
limit. - View the metadata of the retrieval results to confirm that only the knowledge base content corresponding to the currently bound
collectionIdsis included, with no cross-domain documents. - Test questions using different terms, such as "grid connection" and "grid connection procedures", to confirm that the retrieval results are consistent, verifying that
datasetSearchUsingExtensionQueryis working. - After enabling
usingReRank, compare the result ranking before and after enabling to confirm that core process steps appear at the top of the results.
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 2026-09-28.