What the data for this category looks like
Data comes from original records and output documents across the full oil and gas exploration and development workflow. These include single well drilling logs, regional exploration comprehensive reports, trial production data analysis reports, industry compliance documents, and project contract templates. Update frequency adjusts based on project progress. Corresponding well data is updated within 1 to 3 days after new drilling operations finish. Policy-related documents are updated to align with regulatory requirements. Documents contain dedicated fields such as well ID, block number, test pressure, and oil production volume. Most units use industry standard measurements including meters, cubic meters, and tons. Some long documents exceed 10,000 words, with mixed structured tables and unstructured text content.
What constraints these characteristics impose on the knowledge base retrieval and recall workflow
Dedicated fields and units require retrieval to support precise filtering by fields like well ID and block number. This prevents return of documents from unrelated well sites. High proportions of long documents lead to large numbers of single-document chunks. Chunk granularity must be controlled to avoid retrieval redundancy. For project scenarios with high real-time requirements, incremental recall must be supported to quickly access the latest operational data. Industry documents have high densities of professional terminology. The retrieval model must be adapted to semantic encoding of oil and gas exploration-specific terminology, to avoid semantic matching deviations. Mixed structured and unstructured content requires retrieval to support both text semantic matching and structured field matching. This prevents missing key numerical information.
How to configure settings
Reference only; verify against the target FastGPT release and deployment documentation.
| Configuration Item | Recommended Value Range | Rationale |
|---|---|---|
similarity | 0.72–0.85 | Matches the semantic matching accuracy needed for oil and gas industry professional terminology, filters low-correlation results while retaining retrieval of professional documents |
limit | Top 8–12 results | Controls the number of single retrieval recall results, balances information richness and context window usage, and adapts to large model processing limits |
searchMode | Hybrid retrieval | Adapts to mixed structured fields and unstructured text in documents, matches both semantic and dedicated field tags |
usingReRank | Enabled | Performs secondary ranking on initial recall results, corrects deviations in semantic matching of professional terminology, and improves result accuracy |
datasetSearchUsingExtensionQuery | Enabled | Generates oil and gas exploration-specific extended queries based on the original question, supplements synonyms and term variants, and improves recall coverage |
collectionIds | Configured by project/well site group | Limits retrieval scope to the knowledge base collection of the corresponding project, avoiding recall of unrelated cross-project documents |
Three common errors
- Phenomenon: Assistants under different accounts return identical document sets, without distinguishing content across permission scopes. Cause: The
collectionIdsparameter is not used to bind the dedicated knowledge base group for the corresponding account, and permission filtering logic is not configured. - Phenomenon: The number of retrieval results is far lower than the configured
limitvalue, or empty results are returned. Cause: ThecollectionIdsrange is not correctly limited, or thesimilaritythreshold is set too high, filtering most relevant professional documents. - Phenomenon: Generated document content does not reference retrieved professional documents, with generic responses generated directly. Cause:
datasetSearchUsingExtensionQueryis not enabled, causing retrieved professional terminology to not be correctly associated with the generation process by the model.
How to confirm configurations are properly set
- Input dedicated queries containing well IDs and block numbers, check the scope and relevance of retrieval results, and confirm retrieval rules match configured requirements.
- Adjust the
similaritythreshold, observe changes in the number and relevance of retrieval results, and verify the effectiveness of the semantic matching logic. - Enable
usingReRank, compare retrieval results before and after activation, and confirm the reranking function operates normally. - Input generic queries, check whether retrieval results filter unrelated non-oil and gas exploration documents, and confirm the
searchModeconfiguration is effective.
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.