What this category of data looks like
Coal mine production business processing data mainly comes from internal safety regulations, post operation manuals, equipment operation and maintenance ledgers, daily inspection records, and emergency response plans. Regulatory documents are revised annually or quarterly. Inspection and operation and maintenance records are updated daily. Most documents are structured as chaptered standard text, or ledger-style files with standardized fields. These fields include working face number, ventilation parameters, equipment model, fault level, and similar. Common units are meters, hours, MPa, tons, cubic meters per second, and similar.
What constraints do these characteristics impose on the knowledge base retrieval and recall link?
The fixed chapter structure of regulatory documents requires retrieval to match the chapter number corresponding to the process node. This avoids retrieving invalid content across process links. Standardized ledger fields require retrieval to support filtering by fields such as working face number and equipment model. This accurately locates content for the corresponding business scenario. Frequently updated inspection records require the retrieval pipeline to support low-latency recall. This prevents business processing waiting timeouts. Homogeneous records of the same type of operation require recall to distinguish priority based on business scenarios. This prevents irrelevant content from interfering with process guidance.
How to set the configuration
Reference only; verify against the target FastGPT release and deployment documentation.
| Configuration Item | Recommended Value | Rationale |
|---|---|---|
similarity | 0.75–0.85 | Coal mine production business process texts are mostly rigorous and standardized. This range filters low-correlation non-process content while retaining recall results for core steps |
limit | Top 8 entries | The steps of a single business processing flow usually do not exceed 7. Retaining 8 entries covers backup processes and exception handling steps, avoiding excessive results interfering with guidance |
searchMode | semantic | Business process guidance requires combining semantic understanding of step logic. It adapts to colloquial questions from post personnel, and covers semantic association scenarios that keyword matching cannot reach |
usingReRank | Enabled | For homogeneous inspection records and regulatory content, re-ranking adjusts sorting priority and improves the accuracy of process steps |
datasetSearchUsingExtensionQuery | Enabled | Coal mine business terms are mostly industry-specific. Extended queries supplement synonyms and term associations, improving the recall rate of niche processes |
collectionIds | Bind to coal mine-specific process knowledge base collections | Ensure the retrieval scope is limited to the current coal mine production business domain, avoiding retrieving knowledge base content from unrelated scenarios |
Three common mistakes
- Symptom: After importing knowledge base data using the mongorestore tool, no matching results are returned during retrieval, or an empty list is returned. Cause: The target knowledge base collection was not correctly bound during import, or the data metadata did not carry the identification field corresponding to
collectionIds, resulting in the retrieval scope not covering the imported data. - Symptom: When there is a large amount of homogeneous content in the knowledge base, older documents are ranked before newer ones, which does not meet the requirement of date-weighted priority. Cause:
usingReRankwas not configured, or no time-based sorting logic was added. The default recall sorting only relies on similarity and does not distinguish document update times. - Symptom: After performing a knowledge base reindex operation, some documents cannot be recalled normally. Cause: Not all document chunks were fully traversed during reindexing, or the operation was interrupted midway, resulting in indexes for some chunks not being generated. A full index rebuild needs to be triggered again.
How to confirm the configuration is correct
- Import a single test document containing coal mine-specific process keywords. Perform retrieval after importing using the mongorestore tool, and confirm that the test document appears in the recall results.
- View the retrieval log to confirm that the values of configuration parameters such as
similarityandlimitmatch the preset values. - Upload two homogeneous test documents with different upload times. Perform retrieval, and confirm that the newer document is ranked higher.
- Switch
searchModeto other modes, compare the differences in recall results, and confirm that the semantic retrieval mode is active.
The values provided 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.