What the Data for This Category Looks Like
Data for basic chemicals declaration materials and file review comes primarily from internal safety evaluation reports, environmental assessment submission documents, work safety files, raw material quality inspection archives, and process parameter documents. There are two types of data updates: batch upload of temporary data during the project declaration phase, and monthly updates of archived files. Single document lengths vary widely, ranging from hundreds of words of approval opinions to tens of thousands of words of complete declaration plans. Fixed document fields include project number, raw material specification, process parameters, and approval document number. Supporting units mostly use industrial standard units such as ton, cubic meter, and MPa.
How These Characteristics Impact Knowledge Base Retrieval and Recall
The wide range of document lengths for basic chemicals declaration materials and files creates significant differences in context weight for single retrieved content. Adjust the matching granularity of chunking and recall to address this. There are many fixed fields with industrial-specific units. General retrieval models are prone to unit matching deviations. Strengthen field-level precise recall to fix this issue. The update rhythm of batch-uploaded temporary data and archived data differs. Specify the corresponding collection range during retrieval to avoid mixing expired file data. Nested attachments in long documents, such as text descriptions of process flow diagrams, are easily missed. Enable cross-attachment associated retrieval to capture these items.
Configuration Settings
Reference only; verify against the target FastGPT release and deployment documentation.
| Configuration Item | Recommended Value | Rationale |
|---|---|---|
similarity | 0.72–0.80 | Matches the precision of basic chemicals terminology, avoids introducing irrelevant files with too low a value, and avoids missing compliant declaration materials with too high a value |
limit | Top 8–12 results | Balances the coverage of retrieval recall and context length, adapts to the content volume of long basic chemicals documents |
searchMode | semantic + keyword | Balances semantic matching of professional terminology and precise keyword matching of fixed fields such as project number |
usingReRank | Enabled | Corrects sorting deviations of professional parameters in basic chemicals documents, improves the display priority of highly relevant content |
datasetSearchUsingExtensionQuery | Enabled | Automatically expands terms such as raw material names and process parameters in declaration materials, covering synonyms and industry aliases |
collectionIds | Specified by archived collection | Distinguishes temporary declaration data from historical files, avoids recalling expired or non-target category documents |
Three Common Configuration Mistakes
- Phenomenon: A 400 status code is returned when calling the API to add or delete collection data, with a prompt indicating incorrect parameter format. Cause: The file path and metadata fields are not transmitted in the JSON structure required by FastGPT.
- Phenomenon: No results are returned for the first knowledge base retrieval attempt, but matching content is returned on the second attempt. Cause: The vector database has not finished generating indexes for newly uploaded data when the first retrieval is initiated.
- Phenomenon: After passing HTTP response data as a reference, the retrieval result does not associate this content. Cause: The response data is not converted into structured text that meets the knowledge base requirements, and the
datasetSearchUsingExtensionQueryconfiguration for field association extension is not enabled.
How to Verify Correct Configuration
- Initiate a simulated declaration material query, check whether the collection range of the returned results matches the specified
collectionIds. - Adjust the value of the
similarityparameter, verify whether the relevance of the retrieval results meets business expectations. - Upload a test document with industrial parameters, confirm whether
usingReRankadjusts the sorting position of highly relevant content. - Call the API interface to add a test file data, verify whether the content can be normally retrieved during retrieval.
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.