What This Category of Data Looks Like
This category of data mainly originates from ERP orders and inventory, bills of materials (BOMs) and process routes, equipment capacity, staff shifts, work-in-progress inventory, and supplier deliveries. Daily documents include sales orders, master production schedules, material requirement plans, weekly and daily scheduling tables, work orders, material shortage lists, and delivery change notices. Data is updated daily, weekly, or triggered by orders or abnormal events. Document fields include order numbers, material codes, capacity units, shift durations, and more. Different document types have dedicated fields and units.
How These Characteristics Impact the Knowledge Base Retrieval and Recall Link
Multi-source and dynamically updated data requires retrieval to support near-real-time synchronized datasets, to avoid recalling expired scheduling information. Complex document types and field units require retrieval to distinguish documents by business collections, to avoid interference from cross-category data. Core business workflows focus on work order decomposition and resource-constrained scheduling, requiring recalled content to prioritize association with work orders, material requirements, and capacity data. Rapid rescheduling needs in abnormal scenarios require retrieval to quickly locate historical records related to material shortages and ad-hoc order insertions.
How to Configure the Settings
This configuration is adapted for FastGPT v4.17.0.
| Configuration Item | Recommended Setting | Rationale |
|---|---|---|
similarity | 0.75-0.85 | Production planning documents are highly specialized, so low-match irrelevant content must be filtered to avoid interfering with scheduling decisions |
limit | Top 8 entries | The volume of production planning-related documents is concentrated; too many results will exceed the context window and reduce AI processing efficiency |
collectionIds | IDs of collections corresponding to production plans, scheduling tables, and work orders | Distinguish different business document collections, and only recall knowledge base content directly related to shop scheduling |
usingReRank | Enabled | Re-rank the multiple initially recalled results to prioritize content most relevant to the current scheduling task |
datasetSearchUsingExtensionQuery | Enabled | Production planning questions often require supplementary associated material and capacity information; extended queries can improve retrieval accuracy |
searchMode | embedding | For production planning terminology, semantic retrieval better matches query intentions in business scenarios |
Three Common Configuration Errors
- Phenomenon: When submitting questions unrelated to production planning, the knowledge base retrieval branch is still triggered, returning unexpected content. Cause:
collectionIdsis not configured correctly, and retrieval is not limited to production planning-related collections, resulting in cross-collection recall of irrelevant documents. - Phenomenon: Retrieval results contain a large amount of low-match content, and irrelevant information cannot be filtered. Cause: A reasonable
similaritythreshold is not set, orusingReRankis not enabled, causing low-relevance content to be included in the result set. - Phenomenon: The number of retrieval results does not match expectations, with too many exceeding the context window or too few failing to cover full business information. Cause: The
limitparameter is not adjusted based on the total volume of business documents, and the value does not match the actual data scale.
How to Confirm the Configuration Is Correct
- After uploading production planning-related documents, check if
collectionIdsis associated with the IDs of the corresponding business collections, to ensure only documents within that collection are included in the retrieval scope. - Submit a test question such as "This week's machining shop scheduling table", and verify whether the matching score of the returned results falls within the preset
similarityrange. - Submit a test question unrelated to production planning, and confirm that the knowledge base retrieval branch is not triggered, or that the preset designated reply logic is activated.
- Test the multi-document recall scenario, and confirm that after
usingReRankis enabled, the result ranking prioritizes content most relevant to the current scheduling task.
The values provided are common starting points and should be evaluated using samples appropriate to the specific use case.