What This Category of Data Looks Like
The data for this category is generated by manufacturing enterprise change management teams. Daily documents include ECN/ECR, change assessment reports, drawings and BOM version files, process verification reports, transition plans, inventory disposal documents, customer approval documents, and similar items. Data sources cover PLM, ERP, MES, supplier materials, verification test records, and customer special requirements. The update frequency follows an irregular schedule triggered by each change. Document structure includes fields such as change number, version number, effective date, and involved materials/processes/inspection items. The version number uses the standard unit of V + numeric sequence. Dates follow the ISO 8601 format.
Constraints on Knowledge Base Retrieval and Recall
The characteristics of this category of data impose clear constraints on the retrieval and recall process. Multi-source heterogeneous data sources require retrieval to cover cross-system datasets. Omitting drawing versions from PLM or material association information from ERP will lead to data gaps in ledger summaries. The irregular update schedule requires retrieval configurations to support incremental synchronization. This ensures that the latest documents are recalled promptly after each change, avoiding use of expired content. Document types include structured tables, engineering drawings, and plain text reports. Retrieval must support both full-text content search and metadata field matching. This ensures that key information such as change numbers and version numbers is accurately identified, preventing retrieval results from deviating from change topics. The number of documents involved in a single change fluctuates greatly. It is necessary to limit the total number of recalled results. This avoids redundant content in ledger summaries that reduces reading efficiency.
Configuration Settings
This configuration is based on FastGPT v4.17.0. Specific configuration items and explanations are listed below:
| Configuration Item | Recommended Setting | Rationale |
|---|---|---|
similarity | 0.75–0.85 | Matches the accuracy requirement for process change-related documents, filters low-relevance daily production documents |
limit | Top 10 entries | Controls the number of single-recall results, adapts to the length requirements of production ledger summaries, avoids returning content exceeding 2000 characters |
searchMode | fullTextRecall | Use full-text retrieval for text content. Field, metadata, batch, and document-type constraints require separate filters and are not enforced by this mode. |
usingReRank | Enabled | Rearranges recalled results by change effective priority, ensures documents with the latest versions are displayed first |
collectionIds | Bind dataset IDs corresponding to PLM, ERP, and MES | Covers data sources across the entire change management lifecycle, ensures retrieval scope includes all relevant documents |
datasetSearchUsingExtensionQuery | Enabled | Expands the synonym range of retrieval terms, covers related expressions such as "process change" and "engineering change", improves recall coverage |
Three Common Misconfigurations
- Phenomenon: Empty results are returned during retrieval, or a prompt indicates that external data source content cannot be loaded. Cause: Dataset IDs corresponding to data sources such as PLM and ERP are not bound to the current knowledge base collection, or cross-data source retrieval configuration is not enabled.
- Phenomenon: A `
Validation failed: name: Pathnameis required.` error appears when generating a report. Cause: Metadata field mapping during retrieval is not configured correctly, and required fields such as change number and version number are not included in the retrieval scope. - Phenomenon: Expired version process documents are mixed in retrieval results, and results are not sorted by effective time. Cause: The
usingReRankconfiguration is not enabled, or thesimilarityvalue is set too low, leading to the recall of irrelevant results.
How to Verify Successful Configuration
- Access the knowledge base management interface of FastGPT v4.17.0, check if the bound datasets include change-related data sources such as PLM, ERP, and MES.
- Initiate a test retrieval, enter "ECN change number", verify if the returned results include matching content for the corresponding metadata fields.
- Generate a simulated change ledger report, check if results are sorted by version priority and no expired documents are included.
- Review retrieval logs, confirm that the number of recalled results for each query matches the value set in the
limitconfiguration.
The configuration values listed above are common starting points. Adjust them based on your own specific samples and use cases.