What the data for this category looks like
The internal office and employee service data for coal transportation and sales comes primarily from transportation and sales scheduling ledgers, coal quality test reports, purchase and sales contract documents, employee job manuals, and daily office notices. Data update rhythms cover multiple dimensions:
- Real-time scheduling data such as shipment volume and train trips, updated daily
- Single-mine coal quality parameter data, updated per batch
- Employee training materials and contract archive data, updated regularly
Document structures include structured tables (such as ledgers with fields like shipment tons, calorific value in large calories, customer numbers), long-text notices, and job operation guides. Most fields are tied to clear physical units, including tons, large calories, and train trips.
Constraints Imposed on Knowledge Base Retrieval and Recall
Data sources with multiple update rhythms require the knowledge base to support incremental synchronization and batch updates. This avoids full re-imports that consume system resources. A mixed content structure of structured and unstructured documents requires retrieval to balance semantic matching and exact field matching. This prevents omission of precise entries within structured tables. Clear fields and units require retrieval to support scope limitation by business dimensions. Examples include only retrieving that month’s shipping ledgers or coal quality test data for specified mines. This prevents irrelevant cross-business line data from appearing in results. Spoken queries in employee service scenarios require retrieval optimizations. These convert daily expressions into precise keywords that match business documents.
Configuration Settings
| Configuration Item | Recommended Value | Rationale |
|---|---|---|
similarity | 0.72–0.80 | Structured field matching for coal transportation and sales requires a high threshold to avoid irrelevant coal quality parameters or historical contracts from mixing into results |
limit | Top 10–15 entries | Retrieval results for internal office and employee services do not need excessive volume, avoiding information overload while covering major business document types |
searchMode | hybrid | Balances exact matching for structured tables and semantic recall for long text, adapting to the mixed document structure of coal transportation and sales |
usingReRank | Enabled | Improves ranking relevance for mixed documents, especially for mixed retrieval scenarios of long-text notices and structured tables |
datasetSearchUsingExtensionQuery | Enabled | Optimizes employees' spoken queries, such as converting "today's shipping plan" into keywords matching that day's scheduling ledgers |
collectionIds | Limited by business line, such as collection IDs for transportation and sales ledger databases and coal quality test databases | Limits retrieval scope to specified business data sets, preventing irrelevant data recalls from cross-business lines |
Three Common Configuration Mistakes
- Phenomenon: Unable to limit retrieval to specified imported documents. For example, attempting to retrieve only that month’s coal shipping ledgers but recalling all historical data. Cause: The
collectionIdsparameter is not configured correctly, or the retrieval data set scope is not limited, resulting in recall of all knowledge base content. - Phenomenon: Insufficient returned entries when retrieving structured table data for coal transportation and sales. For example, only 30+ entries are returned out of 50 entries in the ledger. Cause: The
searchModeis not set to hybrid mode, relying only on semantic recall and missing exact matching structured entries.
Reference only; verify against the target FastGPT release and deployment documentation.
How to Verify Correct Configuration
- Manually enter a query known to exist in the specified business data set, and confirm the returned results only include documents from that data set.
- Upload a test coal transportation and sales ledger table, run a retrieval, and confirm the number of returned entries matches the actual entries in the table.
- After enabling
usingReRank, compare the sorting of retrieval results before and after enabling to confirm higher-relevance entries appear first. - Adjust the
similarityparameter, and verify whether the relevance of returned results meets business expectations. Do not set a fixed threshold.
The values provided here are common starting points and should be evaluated against the reader's own samples.
Question material comes from public community discussions. Configuration values are common starting points and should be measured against your own samples. Verified on 2026-09-28.