What the data for this category looks like
The data for this category primarily comes from the national public resource trading platform, natural resource trading systems, property right exchange systems, and state-owned asset supervision systems. Commonly processed documents include transfer announcements, trading guidelines, bid application materials, margin vouchers, transaction confirmation documents, listing result announcements, and property right transfer specifications. Data is updated daily. Field units are adjusted flexibly based on document type. Some documents contain long paragraphs or multi-table structures, with core content focused on business nodes such as ownership, valuation, margin, and transaction confirmation.
What constraints these characteristics impose on the knowledge base retrieval and recall link
High-frequency daily updated data requires the retrieval system to support incremental indexing, to avoid resource consumption from full reindexing. Varied structures across different document types require dividing datasets by document type, to prevent retrieval interference between different business documents. The need for precise matching of core fields requires prioritizing association with key information such as ownership and margin during retrieval, to avoid generic retrieval. Documents vary greatly in length; long documents may contain multiple independent business nodes. It is necessary to ensure retrieved content accurately corresponds to the business process step in the user's query, to avoid irrelevant information interfering with the review process.
Configuration Settings
| Configuration Item | Recommended Setting | Rationale |
|---|---|---|
similarity | 0.75-0.85 | This category requires precise matching of core fields such as ownership and margin, to avoid irrelevant low-similarity results interfering with business reviews |
limit | Top 8 entries | Business process guidance typically requires limited reference content; excessive results increase screening costs for review staff |
usingReRank | Enabled | In scenarios with many long documents and complex structures, re-ranking improves the ranking priority of core content, adapting to retrieval needs for key business information |
datasetSearchUsingExtensionQuery | Enabled | Automatically expands relevant business keywords based on user queries, covering retrieval needs for high-frequency business scenarios such as bid applications and transaction confirmations |
collectionIds | Dataset IDs divided by document type | Distinguish between different datasets for land use rights, mining rights, and state-owned property rights transactions to avoid mixed results from cross-category retrieval |
Three Common Errors
- Uploading documents such as transfer announcements or property right transfer specifications results in failed retrieval, shown as no corresponding entry in the knowledge base index list. The cause is failing to enable the index switch for the corresponding document type or failing to adapt to the long paragraph structure of the document.
- Calling the retrieval API fails to display embedded images such as margin vouchers or transaction confirmation documents in third-party systems, shown as no image external link field in the returned results. The cause is failing to enable document image extraction and external link generation configuration.
- After configuring the reference upper limit, some document chunks exceeding the set length are still retrieved, shown as retrieval results containing content beyond the reference upper limit. The cause is failing to associate the reference upper limit verification with the retrieval and recall link.
How to Verify Proper Configuration
- Upload a standard transfer announcement document, wait 10 minutes, then perform a keyword search for "ownership verification", confirm that the returned results contain relevant field content from the document.
- View the knowledge base index status panel in FastGPT v4.17.0, confirm that documents updated on the current day have completed indexing with no failed entries.
- Test calling the retrieval API, check that the number of returned results matches the
limitsetting, and that the similarity of returned results falls within thesimilaritythreshold range. - Upload a property right transfer specification exceeding 1500 characters, perform a keyword search for "bid process", confirm that the retrieved content corresponds to the bid-related paragraphs in the document.
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.