What the Data for This Category Looks Like
This scenario targets enterprises producing automotive interior, household appliance, and medical plastic parts. Daily documents include injection molding process cards, raw material drying records, mold histories, first-piece inspection reports, patrol inspection forms, and material change and machine cleaning records. Data sources cover resin batch property sheets, real-time injection molding machine parameters, mold temperature machine data, MES system records, and dimensional and appearance inspection results. Updates are triggered by production batches, mold and material change operations, and shift patrols. Document structures include structured parameter fields and unstructured descriptive content. Field units include degrees Celsius, seconds, counts, hours, and others. Some documents must be linked to corresponding production batches and mold numbers.
What Constraints Do These Characteristics Impose on the Knowledge Base Retrieval and Recall Link?
Multi-source heterogeneous data sources require retrieval to support cross-system data association matching, and avoid isolated calls for single-type documents. Frequently updated production data requires the knowledge base to support incremental synchronization, to ensure retrieval results align with current production status. Diverse document types include both structured process parameters and unstructured exception descriptions, so both semantic and keyword retrieval logic must be supported. Injection molding processes have high precision requirements for parameters such as raw materials and mold temperature, and exception handling is linked to production safety. The matching accuracy of retrieval results must be strictly controlled to prevent low-relevance content from interfering with drafting work.
How to Set the Configurations
| Configuration Item | Recommended Value | Rationale |
|---|---|---|
similarity | 0.75–0.85 | Injection molding process parameters and exception descriptions have high matching accuracy requirements. This range filters low-relevance irrelevant documents, and avoids risks of material mixing or mold damage caused by overly low parameter matching |
limit | Top 8 entries | Balances the comprehensiveness of retrieval results and drafting efficiency. Excessive results increase screening costs for process documents and exception reports |
searchMode | mixedRecall | Covers the dual needs of structured parameter queries and unstructured exception description queries, and adapts to diverse retrieval intentions in this scenario |
usingReRank | Enabled | Improves the sorting priority of high-matching results, and ensures that exception reports first display solutions that best fit the current issue |
datasetSearchUsingExtensionQuery | Enabled | Supplements synonym matching for injection molding industry-specific terms, such as associating "flash" with "overflow" and other expressions, to expand retrieval coverage |
collectionIds | Divide collection IDs by business module | Limits the retrieval scope to corresponding modules such as process parameter library, exception case library, and mold history library, and avoids interference from irrelevant data across modules |
Three Common Misconfigurations
- Phenomenon: After mounting multiple knowledge bases, only matching results from a single knowledge base are returned. Cause: The
collectionIdsparameter is not configured correctly, and only the collection ID of a single knowledge base is specified, which limits the retrieval scope. - Phenomenon: After importing knowledge base documents via mongorestore, the imported document content cannot be queried within the platform. Cause: The imported documents have not completed vector generation, or the import format does not meet platform requirements, and the index update of the corresponding knowledge base is not triggered.
- Phenomenon: The same problem has matching content in two knowledge bases, but the returned results have no obvious priority sorting. Cause:
usingReRankis not enabled, or a reasonablesimilaritythreshold is not set, resulting in no valid basis for sorting retrieval results.
How to Verify That the Configuration Is Correct
- Log in to the knowledge base management page of FastGPT v4.17.0 platform, check whether the
collectionIdsparameter includes the collection ID of the target business module, and ensure that the retrieval scope covers the required document libraries. - Initiate a test query, enter injection molding exception keywords such as "shrinkage", check whether the number of returned results matches the set value of
limit, and verify the accuracy of the number of retrieved entries. - Compare query results before and after enabling
usingReRank, confirm whether high-matching exception solution documents are displayed first, and verify the activation status of the reranking function. - Enter industry synonyms such as "short shot", check whether documents corresponding to "insufficient filling" can be retrieved, and verify whether the extended query function of
datasetSearchUsingExtensionQueryis effective.
The values provided are common starting points and should be evaluated against the reader's own samples.