What the data for this category looks like
This category covers full-process records for manufacturers of automotive interior, home appliance, and medical plastic parts. Daily documents include injection molding process cards, raw material drying records, mold histories, first-piece inspection forms, patrol inspection logs, and material change and machine cleaning records. Data sources include resin batch and property sheets, injection molding machine operating parameters, mold temperature controller data, MES systems, and dimensional and appearance inspection results. New records are generated per production batch, each mold or material change operation, and at the end of each shift. Each document includes fields such as process parameters, raw material batch number, mold ID, and inspection results. Some records include numeric parameters and associated units.
What constraints these characteristics impose on knowledge base retrieval and recall
The high-frequency updates and multi-type document structures of this category require the retrieval link to support time range filtering at the batch and mold change level. A high proportion of numeric data such as resin properties and injection molding parameters requires retrieval modes that support numeric range matching. Large differences in fields across different documents require precise recall by document type, such as process cards and patrol inspection records. High-frequency updates at the shift and batch level require the retrieval link to support low-latency responses, to avoid increased retrieval time caused by excessive document increments. Risks such as material mixing and mold damage are associated with multiple documents, so cross-document associated recall must be supported to ensure retrieval results cover complete relevant information.
Configuration Settings
This configuration is set based on the knowledge base retrieval function of FastGPT v4.17.0.
| Configuration Item | Recommended Setting | Rationale |
|---|---|---|
similarity | 0.72–0.85 | Matches the precision requirements for numeric content such as injection molding parameters and inspection results, and avoids recalling irrelevant process cards or patrol inspection records |
limit | Top 8 entries | Covers multiple documents associated with a single production batch, while controlling total token consumption per retrieval round |
searchMode | mixedRecall | Supports text matching and numeric range matching, adapting to the retrieval needs of multi-type documents in this category |
usingReRank | Enabled | Performs secondary sorting on recall results, prioritizing mold histories and inspection records that are strongly relevant to production ledger summarization |
datasetSearchUsingExtensionQuery | Enabled | Automatically expands retrieval keywords, adding industry-specific terms such as injection molding process jargon and raw material model numbers to improve recall coverage |
collectionIds | Configured by document type | Separates collections such as process cards and patrol inspection records to enable precise targeted retrieval |
Three Common Misconfigurations
- Phenomenon: In simple workspace knowledge base applications, only 1 document is returned per retrieval, and the interface displays "Knowledge Base Reference (1 entry)". Cause:
collectionIdsgrouping retrieval is not configured by document type, or thelimitvalue is set too low to cover multiple records associated with a single production batch. - Phenomenon: Retrieval response takes too long, and background logs show high token consumption per retrieval round. Cause:
usingReRankis not enabled, and thelimitvalue is too large, which recalls excessive low-relevance historical records and causes context bloat. - Phenomenon: When uploading a single injection molding process card or mold history file larger than 100MB, the interface displays a file size limit prompt and upload fails. Cause: The large document automatic splitting function is not enabled, which does not adapt to the upload requirements of large documents in this category.
How to Confirm Proper Configuration
- Manually retrieve preset production batch keywords, and check whether the number of returned documents matches the expected range of documents associated with a single batch.
- View retrieval logs to confirm that the accuracy of
similaritymatching results meets business requirements, with no large number of irrelevant documents recalled. - Upload a single test document larger than 100MB, confirm that the automatic splitting function works normally, and no upload failure prompt appears.
- Simulate multi-document associated retrieval, confirm that the returned results cover multiple types of documents such as process cards, patrol inspection records, and mold histories.
The values given are common starting points and should be measured against the reader's own samples.