What the data for this category looks like
Business data for internal office and employee services at state-owned and public enterprises in the coal chemical industry mainly comes from safety operation procedures, equipment operation and maintenance manuals, production dispatch logs, employee training courseware, coal quality inspection reports, and similar sources. Update cycles vary across data types: safety regulations and equipment manuals have longer adjustment cycles following technical renovations and regulatory updates; production dispatch logs are updated daily; training courseware is updated quarterly. Documents include hundreds-page long text procedures, structured equipment ledgers (including equipment numbers, operating parameters), unstructured files in PDF, Word, and Excel formats, with many fields containing professional parameters such as operating temperature, maintenance cycle, and ash content index.
What constraints these characteristics impose on the knowledge base retrieval and recall link
The presence of long-text procedures requires that context association be retained during chunking to avoid splitting professional operation steps. The mix of structured ledgers and multi-format documents requires retrieval to support both semantic matching and field association to accurately match equipment parameter-related content. Differences in update cycles across documents require flexible configuration of incremental and full updates to adapt to the update frequencies of different data. Content dense with professional terms requires retrieval to accurately match professional semantics, avoiding irrelevant general results interfering with internal office and employee service query needs.
How to set the configurations
Reference only; verify against the target FastGPT release and deployment documentation.
| Configuration Item | Recommended Setting | Rationale |
|---|---|---|
similarity | 0.72–0.85 | Coal chemical documents are mostly professional technical content, so a high similarity threshold is needed to filter irrelevant results and avoid incorrect matches related to safety and operation and maintenance |
limit | Top 6–10 results | Internal office involves multiple types of documents, and an appropriate number of recalled results covers reference needs for different scenarios while avoiding result overload |
searchMode | Hybrid Retrieval | Balances full-text semantic matching and structured field matching, adapting to the structure of coal chemical documents that include both text descriptions and parameter indicators |
usingReRank | Enabled | Semantic similarity of professional documents is easily disturbed by terms, and re-ranking can optimize the accuracy of result sorting and improve the matching priority of professional content |
datasetSearchUsingExtensionQuery | Enabled | Internal employee questions often use colloquial expressions, and extended queries can associate professional terms to improve recall matching accuracy |
collectionIds | Bind the corresponding coal chemical knowledge base collection ID | Isolate documents from different business lines, avoid interference from irrelevant data on retrieval results, and ensure the security of internal office data |
Three common mistakes
- Phenomenon: When clicking the details of a retrieval result, the complete content of the original document before chunking cannot be viewed. Cause: The knowledge base configuration does not associate the original file storage path, or the association metadata between the original file and the chunk is not retained during chunking.
- Phenomenon: After uploading a coal chemical PDF equipment manual, the retrieval recall results do not include the equipment parameter chapter content. Cause: The
similaritythreshold is set too high, filtering chunks that contain professional parameters but have slightly lower semantic matching scores. - Phenomenon: There is no interactive response after the agent editing page loads, and retrieval parameters cannot be configured. Cause: The knowledge base collection corresponding to
collectionIdsis not correctly specified, causing the page to fail to load associated knowledge base data.
How to confirm the configuration is complete
Submit a coal chemical professional question, such as "What is the maintenance cycle of XX model pump", and check if retrieval results include the chunked content of the corresponding document. Adjust the similarity threshold, observe the change in the number of retrieval results, and confirm that the threshold takes effect. After enabling usingReRank, compare the result sorting before and after re-ranking to confirm that the re-ranking function works properly. Click the detail link of the retrieval result, confirm that the link allows jumping to the corresponding position of the original file for preview.
The values given 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.