What this category of data looks like
This category primarily covers enterprises engaged in color printing, labels, cartons, flexible packaging, and pharmaceutical packaging materials. Daily documents include pre-press files, imposition sheets, printing process sheets, ink color matching records, first-piece proofing, barcode detection, die-cutting and lamination records. Data sources include customer manuscripts, color management systems, printing press operation data, ink and substrate batch records, and collection results from spectrometers and barcode detectors. Data updates occur per order, per batch, or each time a plate is changed. Document fields include version number, ink color difference value, substrate grammage, barcode verification result, die-cutting knife parameters, and some fields have standardized units.
What constraints these characteristics impose on the knowledge base retrieval and recall link
This category has high data update frequency, and includes strongly associated fields such as version number and batch number. It requires knowledge base retrieval to accurately match the version number and batch information of the current order. It also needs to cover content versions, colors, and packaging structures, which differs from paper making which only provides substrates. Retrieval must cover multi-dimensional parameters such as text content, color difference values, and packaging structures. Data sources include printing press operation data and first-piece proofing records, so retrieval must associate synchronized data from different sources. Errors lead to severe consequences: wrong plate versions, barcode errors, or unqualified packaging performance will cause full batch scrapping. Retrieved results must match the verification rules of the current production link.
How to set configurations
These configurations are based on FastGPT v4.17.0.
| Configuration Item | Recommended Value | Rationale |
|---|---|---|
similarity | 0.85–0.92 | This category requires precise matching of fine fields such as version number and color difference value. This interval balances retrieval precision and effective recall volume |
limit | Top 8 entries | This category's single-order associated documents cover multi-link data such as process, quality inspection, and batch. 8 entries can cover core information |
usingReRank | Enabled | This category has multiple field dimensions. Reranking filters redundant results that do not belong to the current version or batch |
datasetSearchUsingExtensionQuery | Enabled | This category requires associating information between customer manuscripts and production data. Extended queries supplement associated fields such as version number and batch |
searchMode | mixedRecall | This category matches both text content and structured fields such as color difference values and barcode numbers |
collectionIds | Bind three sub-collections: pre-press documents, process records, and quality inspection results | Documents in this category are classified by production link. Binding the corresponding collections narrows the retrieval scope and improves efficiency |
Three common misconfigurations
- Phenomenon: Knowledge base search returns empty results. Cause: The current order's version number and batch information are not included in the extended query, resulting in retrieved documents that do not match current production requirements.
- Phenomenon: Some queries in conversations do not trigger knowledge base retrieval. Cause: Forced retrieval logic is not configured, so the system judges that there is no need to call the knowledge base based on context.
- Phenomenon: After multiple queries in the same conversation window, knowledge base retrieval time increases significantly. Retrieval speed is normal in new windows. Cause: Session context cleaning rules are not configured. Accumulated historical data expands the matching range of vector retrieval, increasing retrieval time.
How to confirm the configuration is correct
- Enter a query that includes a specific version number, check whether the retrieved results include the printing process sheet of the corresponding version.
- Enable the forced retrieval configuration, initiate any production-related query, confirm that the knowledge base is called for each query.
- Initiate more than 10 consecutive production-related queries, compare the retrieval time between new windows and old windows, confirm there is no significant time difference.
- Enter a query with no matching documents, confirm that the retrieval returns empty results as expected.
The values given are common starting points and should be measured against the reader's own samples.