What the data for this category looks like
The data for construction general contracting project declaration materials and file review mainly comes from project review ledgers, construction drawing design documents, construction logs, qualification certification documents, acceptance record forms, and similar sources. Data is updated along with project review nodes. The number of new documents added per review batch fluctuates with project phases. Documents include structured fields such as project number, section name, review date, and unstructured text. Some documents are attached with engineering measurement units including square meters, cubic meters, and ten thousand yuan. Supported formats include PDF, Word, Excel, and public web links.
What constraints these characteristics impose on the knowledge base retrieval and recall link
The data for construction general contracting project declaration materials and file review includes structured fields and multi-format unstructured content. This requires retrieval to match both field keywords and engineering professional terms, with clear requirements for recall accuracy. Project data is updated in batches along with review nodes, so support for incremental recall and filtering by project section dimension is needed. Scenarios where public web links are used as data sources must comply with compliant web parsing rules. For knowledge bases storing multi-format documents, the system must support text extraction logic for different formats to avoid format parsing failures.
How to Configure the Settings
These configurations apply to FastGPT v4.17.0.
| Configuration Item | Recommended Range | Rationale |
|---|---|---|
similarity | 0.75–0.85 | Matches the accuracy requirements for construction engineering professional terms and structured fields, and avoids recalling irrelevant non-engineering documents |
limit | Top 8–12 entries | Adapts to the number of documents per review batch, ensures retrieval results cover main types of declaration materials, and avoids excessive results interfering with the review process |
searchMode | Hybrid retrieval | Supports both structured field keyword matching and semantic recall of unstructured professional text, adapting to the mixed content scenario of construction project review materials |
usingReRank | Enabled | Re-ranks initial recall results by relevance, optimizing the sorting accuracy of engineering professional term matches |
datasetSearchUsingExtensionQuery | Enabled | Automatically expands engineering domain synonyms and professional terms, improving recall rate for cross-term retrieval |
collectionIds | Configured by project section | Limits the retrieval scope to the knowledge base collection of the corresponding section, avoids interference from cross-project documents, and improves review efficiency |
Three Common Misconfigurations
- Uploading declaration material documents in Markdown format fails, and the interface returns a parsing error. The reason is that construction project review materials often contain special symbols and formulas from engineering drawing annotations, and some Markdown parsing logic cannot support non-standard code blocks or drawing tags.
- Importing a publicly shared web link results in a parsing failure prompt. The reason is that dynamic loading content on the web page is not adapted, only static public web pages are supported as data sources. The document must first be exported to a standard format before uploading.
- Deleting large folders categorized by business in the knowledge base results in a
timeout of 60000ms exceedederror. The reason is that the folder contains a large number of construction project review documents, and the default timeout duration is insufficient to complete the batch deletion operation.
How to Verify the Configuration is Correct
- Initiate a query containing construction engineering professional terms, and check whether the
similarityscore of the returned results falls within the configured range. - Check that the number of retrieval results does not exceed the configured
limitupper limit. - Import a public web link, confirm that parsing is successful and the corresponding knowledge base document is generated.
- Perform a batch deletion operation on a small number of documents, confirm no timeout errors occur, then gradually expand the test scope.
The values provided here 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.