What this category of data looks like
Typical data for this subcategory comes from national public resource trading platforms, natural resource trading systems, property right exchange systems, and state-owned asset supervision systems. Data is updated daily. Daily documents include transfer announcements, trading guidelines, bid applications, margin vouchers, transaction confirmation letters, listing result announcements, property right transfer specifications, and similar materials. Document structures vary widely: some are long-paragraph explanatory text, others are form-based structured fields. Fields include ownership numbers, target area, margin amount, trading process nodes, and more. Units match standard measurement identifiers such as land area and transaction amount based on document type.
What constraints do these characteristics impose on knowledge base retrieval and recall?
Daily high-frequency updates require the retrieval system to support incremental synchronous updates. This avoids resource consumption and delays caused by full scans. Differences in document structure require the retrieval system to adapt to multiple chunking strategies. Split form-based documents by fields, and announcement-style documents by paragraphs. This ensures complete recalled content. Precise matching of core transaction information such as ownership details and margin amounts requires strict keyword matching during retrieval. This avoids invalid results from broad matching. High confidence requirements for state-owned asset data require a post-recall verification step. This ensures returned content matches official documents exactly.
How to set configurations
Reference only; verify against the target FastGPT release and deployment documentation.
| Configuration Item | Recommended Value | Rationale |
|---|---|---|
similarity | 0.75–0.85 | Core content such as ownership information and trading rules needs high matching accuracy to filter low-relevance broad matching results |
limit | Top 6 entries | A moderate number of recalled results per batch prevents context overload that slows response speed and reduces answer accuracy |
usingReRank | Enabled | Runs secondary ranking on initial recall results to prioritize content with higher matching degrees for ownership and target information |
searchMode | Precise matching mode | Meets the structured field and precise keyword needs of transaction documents, avoiding false recalls from fuzzy matching |
datasetSearchUsingExtensionQuery | Disabled | Service consultation must strictly use official document content. Generative expansion that deviates from official statements is prohibited |
Three common configuration errors
- Retrieval results fail to match transaction documents in the local knowledge base. This occurs when
collectionIdsis not correctly configured to bind the corresponding public resource trading dataset, or thesimilaritythreshold is set too high, filtering valid matching results. - Some PDF documents cannot be recognized or imported into the knowledge base. This happens when documents contain encrypted content, scanned images, or non-standard form structures, and OCR recognition or document-type-specific chunking rules are not enabled.
- Retrieval responses time out or return abnormal result counts. This is caused by not restricting the
limitparameter, leading to excessive document loading and processing, or a mismatch between the query model and embedding model’s computing power, which increases processing time.
How to confirm the configuration is correct
- View the knowledge base collection management page in the FastGPT backend, and confirm that the
collectionIdsparameter is bound to the exclusive dataset for land use rights, mining rights, and state-owned property rights transactions. - Input precise keywords such as ownership numbers and margin amounts for testing, and verify that retrieval results match corresponding content in official documents.
- Upload different types of transaction documents, including transfer announcements and bid applications, and verify that the system correctly recognizes and chunks all document structures.
- Simulate high-frequency query scenarios, test response speed and result completeness, and confirm there are no timeouts or missing results.
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.