What this category of data looks like
Reference only; verify against the target FastGPT release and deployment documentation.
What constraints do these characteristics impose on the knowledge base retrieval and recall link
The characteristics of document data for general contracting of building construction projects impose multiple constraints on the retrieval and recall link. First, wide variation in document length requires adaptive chunking and recall rules to avoid losing critical context in long documents or over-splitting short documents. Second, structured fields such as project number and section require filtering by project dimension during retrieval to avoid interference from irrelevant cross-section results. Third, dense technical terms such as CFG pile, post-pouring strip, and post-pouring strip water stop require both semantic matching and precise keyword recall. Fourth, update rhythm fluctuates with project nodes, requiring support for incremental sync and real-time recall of the latest visas and acceptance documents.
How to configure the settings
| Configuration Item | Recommended Setting | Rationale |
|---|---|---|
Similarity threshold (similarity) | 0.72–0.85 | There are many construction-specific terms, this range filters low-relevance results and avoids confusion between similar construction process descriptions |
Single search limit (limit) | Top 10 entries | Construction documents are archived by section, relevant results for a single section are concentrated within 10 entries; too many results exceed context capacity limits |
Search mode (searchMode) | Hybrid vector + keyword retrieval | Construction documents have both structured fields and unstructured professional content, hybrid mode balances precision and recall, covering keyword matching for titles and main text |
Rerank toggle (usingReRank) | Enabled | Semantic connections between construction-specific terms are complex, reranking corrects semantic bias from vector retrieval and improves ranking priority of relevant professional results |
Query optimization toggle (datasetSearchUsingExtensionQuery) | Enabled | Project document queries often omit project numbers; optimization automatically supplements project dimension keywords to improve recall precision for documents in the same section |
Dataset filter ID (collectionIds) | Bind the knowledge base collection ID for the current section | Construction projects are archived independently by section; filtering only recalls documents for the current project to avoid irrelevant cross-section content interfering with retrieval results |
Three common mistakes
- When uploading documents to sync with the knowledge base, "Request failed with status code 400" is returned. The cause is that the section knowledge base collection corresponding to
collectionIdsis not correctly bound, or the uploaded visa or acceptance report documents have non-standardized field formats, triggering interface parameter verification failure. - When configuring custom chunking rules, the ideal chunk length is set to a fixed value, which does not adapt to the varying lengths of construction documents. The cause is that the feature of construction projects having both tens-of-page general contracts and single-page construction logs is not considered; uniform chunking will lose critical context association information.
- Knowledge base retrieval results do not include document title-related content and responses are slow. The cause is that
searchModeis only set to pure vector retrieval, and hybrid retrieval mode is not enabled, andlimitis set to more than 15 entries, causing the system to load too many low-relevance document fragments.
How to confirm the configuration is correct
- Upload a single section-specific construction log document, trigger sync, then check the knowledge base sync log to confirm there are no 400-level errors and the document has been correctly indexed.
- Enter a search term that includes a project number, check the filtering effect of retrieval results to confirm only documents for the current section are returned.
- Enable
usingReRank, then compare retrieval results before and after enabling to confirm the ranking of results related to professional terms has improved. - Enter a search term that includes document title keywords, check whether retrieval results preferentially return documents for the current section to confirm the title retrieval function is working properly.
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.