What this category of data looks like
This category of data primarily originates from the engineering quality and safety supervision system, construction site real-name management platform, supervision logs, and test reports. Daily documents include supervision plans, supervision inspection records, hidden danger rectification notices, quality problem handling forms, and completion acceptance supervision reports. Data is updated weekly. Most individual documents contain structured fields such as on-site inspection items, rectification requirements, acceptance verification conditions, and status identifiers. Some documents are semi-structured content transcribed from handwritten on-site scanned documents.
What constraints do these characteristics impose on knowledge base retrieval and recall
The data characteristics of this category create multiple constraints for knowledge base retrieval and recall. Weekly updated data requires the system to support flexible configuration of incremental synchronization and full refresh to maintain data timeliness. Mixed semi-structured and unstructured document types require support for multi-modal retrieval and precise field-level recall. Segmented content such as on-site inspection items and rectification requirements in individual documents requires retrieval to match business scenario keywords, to avoid generalized recall. Content transcribed from on-site scanned documents has format errors, so fault-tolerant retrieval rules must be configured to reduce false recall rates.
How to set configurations
| Configuration Item | Recommended Setting | Rationale |
|---|---|---|
similarity | 0.72–0.78 | Meets the accuracy requirements for policy clause matching in business scenarios, and avoids mixing low-relevance results |
limit | Top 8 entries | Covers the typical number of associated contents in a single supervision document, balancing recall scope and context length |
searchMode | Keyword + vector hybrid retrieval | Adapts to document types with mixed structured fields and unstructured text |
usingReRank | Enabled | Performs secondary sorting on recall results to improve the display priority of highly relevant content |
datasetSearchUsingExtensionQuery | Enabled | Automatically generates extended retrieval terms based on user questions to cover synonymous expressions of policy clauses |
collectionIds | Bound by business collection | Only retrieves the knowledge base collection exclusively dedicated to construction engineering quality and safety supervision |
Three common misconfigurations
- Uploading supervision inspection records and rectification notice data in Excel format leads to poor retrieval performance. This occurs because field-level indexes are not configured for structured fields within the table, and only plain text content is vectorized as a whole, resulting in insufficient keyword matching accuracy.
- Incorrect configuration of the
collectionIdsparameter causes retrieval results to include irrelevant documents from other business collections. This occurs because the exclusive knowledge base collection is not bound according to business scenarios, so the system defaults to retrieving all data. - In FastGPT v4.17.0, multi-retrieval-term recall results only display a single user question entry. This occurs because this version does not optimize the log display logic for multi-retrieval terms. This does not impact actual retrieval effectiveness, but reduces debugging efficiency.
How to verify correct configuration
- Upload a single supervision plan document, initiate a query containing specific inspection items, and confirm that returned results only include documents within the current business collection.
- Adjust the
similarityparameter value, compare the number of recall results across different thresholds, and confirm that matching accuracy meets business expectations. - Enable
usingReRankand run a test, confirm that the sorting of returned results aligns with business relevance. - View the retrieval log, confirm that extended retrieval terms generated by
datasetSearchUsingExtensionQuerycover synonymous expressions of the query.
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.