What This Category of Data Looks Like
Ledger and report data in the survey and design field originates from on-site survey raw records, phased design drawings, project change ledgers, and approval and reply documents. The update rhythm shifts dynamically with project progress. Key stages including preliminary survey, detailed survey, and construction drawing design each trigger data updates. Documents contain structured fields and unstructured content. Structured fields include project number, survey point coordinates, material strength parameters, approval nodes, and units such as meters, megapascals, square meters, and workdays. Unstructured content includes full survey reports and design description PDFs.
What Constraints These Characteristics Impose on Knowledge Base Retrieval and Recall
The multi-field structured nature of survey and design data means retrieval must match both business fields and semantic content. This avoids field matching deviations caused by relying solely on semantic recall. Long documents make up a large share of the dataset. When chunking documents, retain contextual connections between adjacent fields to avoid losing parameter association information after splitting. Data updates adjust dynamically with project stages. Retrieval must prioritize recently updated content to avoid using expired survey or design data. In scenarios with multiple coexisting units, retrieval must link fields to their bound unit relationships to prevent confusion of measurement standards for different parameters.
Configuration Settings
| Configuration Item | Recommended Value | Basis for This Setting |
|---|---|---|
similarity | 0.75–0.85 | Meets the precision requirements for matching survey and design parameters, filters low-relevance non-business documents |
limit | Top 8 entries | Covers multi-field associated information in ledgers, avoids missing parameters caused by too few recalled entries |
searchMode | hybrid | Combines structured field matching and semantic retrieval, adapts to multi-field structured survey and design data |
usingReRank | Enabled | Performs secondary ranking on initial recall results, corrects semantic deviations after long document chunking |
datasetSearchUsingExtensionQuery | Enabled | Automatically expands business terms, for example, associates "compressive strength" with parameter retrieval for "MPa" units |
collectionIds | Dataset ID bound to the project | Complies with FastGPT v4.17.0 dataset restriction rules, limits retrieval scope to the current project’s survey and design ledgers and reports, avoids interference from cross-project data |
Three Common Configuration Mistakes
- Phenomenon: Retrieval results do not match the business question. For example, querying "construction drawing ledger for a specific project" recalls irrelevant survey point records. Cause: Did not use the hybrid retrieval mode of
searchMode, relying solely on pure semantic recall and ignoring structured field matching. - Phenomenon: Semantic retrieval scores show values above 0.9, but results include expired survey data. Cause: Did not adjust dataset update frequency based on project stage, and did not prioritize recalling recently updated content.
- Phenomenon: Units are confused in retrieval results. For example, querying "steel tensile strength (MPa)" recalls parameter records using psi units. Cause: Did not enable
datasetSearchUsingExtensionQuery, and did not link parameters to their corresponding unit retrieval rules.
How to Confirm Correct Configuration
- View the retrieval configuration panel, confirm that
collectionIdsis bound to the current project’s survey and design dataset, with no cross-project datasets included. - Initiate a test query, verify that retrieval results match both business fields and semantic content. For example, when querying "survey coordinates of a specific point", results include the corresponding coordinate values and point descriptions.
- Check the retrieval score and result count, confirm that the
similaritythreshold andlimitcount meet the current project’s business precision requirements. Adjust based on test results. - After enabling
usingReRank, compare results before and after sorting, confirm that secondary ranking corrects semantic deviations after long document chunking.
The values provided are common starting points and should be evaluated against the applicable 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.