What this category of data looks like
This category of data comes from four main sources: communication tower operation and maintenance archives, base station declaration approval dossiers, on-site survey transcripts, and infrastructure ledgers. Updates happen when new base stations go live or operation and maintenance changes take place. Batch synchronization runs alongside quarterly case archiving. Document structure includes standardized declaration forms, operation and maintenance log snippets, case numbers, and approval node records. Standardized forms contain tower height, longitude and latitude, and property information. Fields use clear units: tower height is measured in meters, longitude and latitude use decimal format, and case numbers have fixed business prefixes.
What constraints these characteristics impose on knowledge base retrieval and recall
This category of data mixes structured fields and unstructured text. Retrieval must match both text semantics and structured metadata. This avoids irrelevant results caused by relying solely on semantic recall. Case numbers and tower parameters use fixed formats. Field-level matching logic must be retained during retrieval to prevent losing precise matches due to format differences. Individual case documents vary widely in length. Some are hundreds of characters of approval opinions, while others are tens of thousands of characters of survey reports. Retrieval and recall logic must adapt to different text chunk lengths. Data updates trigger with business changes, leading to batch fluctuations. Incremental synchronization rules must be configured to avoid recalling expired operation and maintenance and approval data.
How to set up the configuration
The following configurations are formulated for FastGPT v4.17.0.
| Configuration Item | Recommended Setting | Rationale |
|---|---|---|
similarity | 0.72–0.80 | Matches the precision requirements for mixed structured parameters and unstructured text, and avoids recalling irrelevant operation and maintenance documents |
limit | Top 8 entries | Balances recall coverage and context length, and meets the material quantity needs for single-round case review |
searchMode | hybrid | Supports both semantic retrieval and metadata field matching, enabling precise recall of structured parameters such as tower height and longitude and latitude |
usingReRank | Enabled | Re-ranks initial recall results, corrects semantic matching deviations, and prioritizes content related to core approval work |
datasetSearchUsingExtensionQuery | Disabled | This category of data has a fixed format. No additional query word expansion is required, to avoid introducing irrelevant interference items |
Three common configuration errors
- Phenomenon: Retrieved text blocks do not carry metadata such as tower number and longitude and latitude, making association with specific base station cases impossible. Cause:
searchModeis not set to hybrid mode. Only pure semantic retrieval is enabled, and metadata field matching logic is not bound. - Phenomenon: Extra spaces appear between tower height values and the "meter" unit in documents recalled from the knowledge base, which does not match the original entry format. Cause: Text normalization rules are not configured, and automatic correction of spaces between numbers and units is not enabled.
- Phenomenon: The title hierarchy and subordinate logic of markdown-formatted survey reports are lost after import into the knowledge base, making association with approval opinions of corresponding chapters impossible during retrieval. Cause: Document structure recognition configuration is not enabled. Only plain text segmentation is used, and the subordinate relationship between titles and content is lost.
How to confirm the configuration is correct
- Submit a test query that includes tower height parameters, and verify that returned results include both semantically matched text and corresponding metadata fields.
- Upload a test document that includes numbers and units, and verify that the correct format is retained between numbers and units in recalled results.
- Upload a markdown test document with hierarchical titles, and verify that retrieval results can associate content under the corresponding titles.
- Trigger an incremental synchronization, and verify that only the latest declaration case data is updated in the knowledge base, with no expired content retained.
The values given 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.