What the data for this category looks like
Coal transportation and sales business data mainly comes from three types of sources. First, daily transportation and sales documents. These include structured documents such as weigh slips, settlement sheets, and transportation contracts. They contain fields like transportation and sales order number, departure location, destination, coal type, settlement unit price, and transportation trip count. Units include tons, yuan per ton, and trips. Second, unstructured documents such as official transportation and sales policies and service guidelines. Third, a Q&A library of past user inquiries. Data updates trigger when contracts change, policies adjust, or inquiry content iterates. There is no fixed update cycle. The length of individual documents varies significantly, ranging from tens of characters for document summaries to thousands of characters for policy files.
What constraints these characteristics impose on the knowledge base retrieval and recall link
Structured documents have clear fields. Retrieval must support exact field matching to prevent irrelevant results from being included. Wide variation in document length requires adaptive chunking strategies. These strategies avoid over-truncating long documents and prevent failed matches for short documents. Non-fixed update cycles require the retrieval system to support incremental index synchronization. This eliminates the need for full index rebuilds every time. A total data volume of hundreds of thousands requires reasonable control of recall scope. This avoids returning too many results and increasing context processing burden, while ensuring retrieval speed.
Configuration Settings
The following configurations apply to FastGPT v4.17.0.
| Configuration Item | Recommended Setting | Rationale |
|---|---|---|
similarity | 0.72–0.85 | Matches the professional term similarity requirements of coal transportation and sales business, filters low-relevance results while covering core business keywords |
limit | Top 10 entries | Adapts to the retrieval load of hundreds of thousands of entries, avoids returning too many results and increasing context processing pressure |
searchMode | Hybrid retrieval | Supports both keyword matching and semantic retrieval, adapts to different retrieval needs for structured documents and unstructured content |
usingReRank | Enabled | Reranks recall results to optimize sorting accuracy for coal transportation and sales professional terms |
datasetSearchUsingExtensionQuery | Enabled | Automatically expands business synonyms, such as expanding "thermal coal" to "steam coal", to expand retrieval recall scope |
collectionIds | Specify the current coal transportation and sales knowledge base collection ID | Limits retrieval scope to avoid interference from irrelevant data across business collections |
Three Common Configuration Mistakes
- Phenomenon: Search returns far fewer results than expected, or no matching results. Cause:
datasetSearchUsingExtensionQueryis not enabled. Failure to cover coal transportation and sales professional synonyms leads to insufficient recall scope. - Phenomenon: Knowledge base files are not immediately synchronized to the index after upload. Cause: Incremental index trigger rules are not configured. Only full index updates are supported, which cannot meet on-demand synchronization business requirements.
- Phenomenon: Uploading files smaller than 4MB succeeds, but uploading a 1MB file fails. Cause: Nginx forwarding restrictions during local deployment override FastGPT's default configuration. The actual upload limit differs from the platform's preset value.
How to Confirm Proper Configuration
- Enter coal transportation and sales professional terms, such as "thermal coal settlement process", check if the similarity value of returned results falls within the 0.72–0.85 range.
- Manually modify a transportation and sales document to trigger incremental synchronization. Check that the modified content can be retrieved after the index updates.
- Upload test files of different sizes to confirm no errors occur during upload and index processes.
- After enabling the reranking function, check if returned results are sorted according to business relevance logic.
The values given are common starting points and should be measured 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.