What data for this category looks like
Refinery and chemical production business data primarily comes from internal enterprise process specifications, equipment operation and maintenance manuals, safety work regulations, emergency response plans, daily inspection records, equipment ledgers, and process parameter ledgers. The update frequency varies by document type: process specifications and safety regulations are updated when revised, daily inspection records are archived and updated monthly, and equipment ledgers and real-time process parameters are updated in real time alongside operation and maintenance activities. Most documents are in structured formats, containing fields such as process number, equipment model, operation steps, safety limits, and use industrial standard units including MPa, ℃, cubic meters per hour, and others.
What constraints these characteristics impose on knowledge base retrieval and recall
Structured process parameters and precise units require that retrieval results must match specific processes and equipment; content with mismatched parameters must be filtered out, as this will affect the accuracy of business processing. The asynchronous update rhythm of multi-source heterogeneous data requires that different types of documents be divided into independent datasets, to avoid mixing retrieval results caused by unsynchronized updates. Long-text operation step documents require that retrieval and recall retain contextual associations, to prevent loss of key process information after splitting. Content dense with professional terminology also requires the retrieval link to have semantic correction capabilities, to filter out low-match irrelevant results.
How to configure the settings
| Configuration Item | Recommended Value | Rationale |
|---|---|---|
similarity | 0.75–0.85 | Refinery and chemical production data includes precise process parameters and professional terminology, requiring filtering of low-match irrelevant results to avoid returning operation guidance for non-corresponding processes |
limit | Top 8 entries | Individual refinery and chemical business documents have lengthy content; too many recall results will exceed the context window. 8 entries can cover core operation steps and parameter requirements |
searchMode | vector | Refinery and chemical business process guidance requires precise matching of operation steps and parameters. Vector retrieval is more suitable for structured professional content than keyword retrieval |
usingReRank | Enabled | Refinery data contains many professional terms. Reranking can correct semantic deviations in initial recall, ensuring returned results are strongly relevant to business processing queries |
datasetSearchUsingExtensionQuery | Disabled | Refinery and chemical business queries are mostly for specific processes or equipment operations. Extended queries easily introduce irrelevant general operation and maintenance content, disrupting precise matching |
collectionIds | Fill in according to the bound refinery-specific dataset ID | Differentiate refinery and chemical production data from other business data, avoiding recalling documents from other categories |
Three common misconfigurations
- Phenomenon: Retrieval results only return fragments of the original input query string, without associating upstream and downstream operation steps. Cause:
datasetSearchUsingExtensionQueryis not enabled, and the query is not reasonably extended, making it impossible to associate relevant pre-operation or post-operation process content. - Phenomenon: Business guidance returned after calling the knowledge base does not match the current query, and involves operation processes for other refinery and chemical equipment. Cause:
collectionIdsis configured incorrectly, recalling content from non-refinery and chemical production datasets, or thesimilaritythreshold is set too low, mixing in low-match irrelevant results. - Phenomenon: No returned results or a 500 status code is returned when calling the knowledge base online. Cause: The correct bound refinery-specific
collectionIdsis not configured, or thelimitis set too high, causing context window overflow and preventing the system from completing the retrieval response.
How to confirm the configuration is correct
- Submit a query containing specific equipment models and process parameters, and check whether the returned results include operation steps and parameter limits for the corresponding process.
- View the retrieval log to confirm that all recall result sources are the bound refinery-specific datasets, with no content from other categories mixed in.
- Test queries of different complexity levels to verify that the match degree of returned results falls within the preset
similaritythreshold range. - Confirm that the system version is FastGPT v4.17.0 or later, to ensure that the index update function for image documents works properly.
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.