What the data for this category looks like
The data for this category primarily comes from the 12345 citizen hotline platform, contractor business systems, official knowledge bases, public feedback and evaluation data, and local hotline management regulations. Daily processed documents include request registration forms, dispatch records, handling responses, follow-up visit records, delay applications, return explanations, and standardized knowledge base entries. Data is updated daily, covering full-process ticket-related information from request receipt to closed-loop follow-up visits.
What constraints do these characteristics impose on the knowledge base retrieval and recall link
Daily updated full-process ticket data requires retrieval to support incremental recall, avoiding performance loss caused by full-scale scans. Multiple document types include structured fields and unstructured text. Recall filtering rules must be set based on dimensions such as request type and contractor unit. Some scanned ticket documents require OCR conversion before they can be effectively retrieved. In scenarios with parallel ticket processing, batch retrieval call logic must be adapted. This ensures single retrieval response speed meets ticket handling time requirements.
How to set configurations
Reference only; verify against the target FastGPT release and deployment documentation.
| Configuration Item | Recommended Value | Rationale |
|---|---|---|
similarity | 0.75–0.85 | Balances precise recall and effective coverage when matching ticket request text with knowledge base entries, adapting to the diversity of 12345 citizen requests |
limit | Top 10 entries | Controls the number of knowledge base entries recalled in a single retrieval, avoiding context overload that impacts response generation efficiency |
usingReRank | Enabled | Re-ranks initial recall results to improve matching relevance with current ticket requests, adapting to complex request scenarios |
datasetSearchUsingExtensionQuery | Enabled | Generates extended query terms based on original requests to cover synonyms and similar expressions, adapting to the colloquial nature of public requests |
searchMode | Semantic retrieval | Matches the semantic intent of tickets, covering semantic association scenarios that keyword matching cannot, adapting to non-standard expressions in 12345 requests |
Three common mistakes
- Retrieval response timeout occurs, with request timeout prompts displayed on the interface. The cause is failure to adjust the recall range based on daily updated ticket data scale. Full-scale knowledge base scans cause excessive resource occupancy.
- Scanned PDF document retrieval results are empty or have very low matching degrees. The cause is failure to perform OCR preprocessing on scanned image-format ticket documents. Valid text cannot be extracted for retrieval.
- Knowledge base text segmentation results are unexpected. Content is not split according to actual line breaks. The cause is incorrect configuration of segmentation rules. The string representation of "\n" is used as the split basis instead of correctly identifying actual line breaks.
How to confirm that configurations are correctly set
- Import a scanned-format ticket document, perform a retrieval test, and confirm valid text is extracted and matching knowledge base entries are returned.
- Submit multiple parallel simulated ticket requests, check that retrieval response times meet expectations, with no obvious abnormal fluctuations.
- After configuring text segmentation rules, import a document containing mixed line breaks, confirm content is split according to actual line breaks.
- Adjust the
similarityparameter range, verify that changes in recall result matching degrees meet expectations.
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.