What the data for this category looks like
Data related to municipal water services inquiry and response primarily comes from water utility work orders for water supply repairs, water quality test reports, pipeline inspection records, citizen inquiry history from government service platforms, and officially published service guidelines. Work order data includes user requests, processing progress, and response content, and is updated in real time. Water quality reports and inspection records are updated weekly or monthly, and include structured fields such as sampling points, test indicators, and test times. Service guidelines are updated quarterly, and include item names, eligibility requirements, required materials, and similar content. The data covers three formats: short text inquiry requests, long document reports, and structured test data.
What constraints these characteristics impose on the vector model and indexing workflow
The data includes short text inquiries, long document reports, and structured test data. This requires the vector model to support embedding processing of multi-format text and avoid embedding bias for structured numeric fields. The update frequencies of different data types vary significantly. Real-time work orders require incremental index updates, while periodically updated reports support scheduled full index updates. This necessitates support for mixed update strategies. Short text inquiries account for a large proportion of the data, so the vector model must deliver stable semantic matching for short queries and avoid invalid retrieval of redundant data. The scale of pipeline and work order data grows with the expansion of service coverage, so indexes must support dynamic scaling to maintain retrieval efficiency.
Configuration Guidelines
Reference only; verify against the target FastGPT release and deployment documentation.
| Configuration Item | Recommended Range | Rationale |
|---|---|---|
| Chunk Length | 800–1200 characters | Adapts to the single-block text length of water work orders and service guidelines, avoids information fragmentation within blocks |
| Overlap Length | 100–150 characters | Preserves context between blocks, prevents workflow documents from being split mid-step |
| Similarity Threshold | 0.75–0.85 | Matches short text inquiries from citizens, filters low-relevance retrieval results |
maxToken | 8192 | Adapts to embedding processing of long documents, covers complete water quality test reports or pipeline inspection records |
| Single Search Limit | Top 10 results | Balances retrieval efficiency and result coverage, aligns with common government inquiry needs |
| Reranking | Enabled, top 3 results | Optimizes result ranking for short queries, improves accuracy of citizen inquiry responses |
Three Common Configuration Errors
- Phenomenon: An error message "This token is not authorized to use the model" is returned when calling the embedding model. Cause: No local embedding model or correct third-party access channel is configured, and an unlicensed remote model interface is used directly.
- Phenomenon: After configuring a third-party channel, FastGPT still prompts that no available embedding channel is available. Cause: The channel's model name does not match FastGPT's mapping rules, or the access key for the channel interface was not filled in correctly.
- Phenomenon: Knowledge base search takes too long, with retrieval result delay exceeding 20 seconds. Cause: Chunk parameters were not adjusted for the text length of water service data, and the number of recalled entries per search was not limited, leading to excessive load on vector database retrieval.
How to Confirm Proper Configuration
- Check the embedding model call logs to confirm that each call returns valid vector data with no permission or channel errors.
- Test short text inquiry queries such as "How to apply for water supply account opening", verify the relevance and ranking logic of retrieval results to confirm alignment with the configured similarity threshold and reranking rules.
- Check the index update task logs to confirm that incremental or full update tasks execute normally according to the preset update schedule.
- Test embedding processing of long documents such as pipeline inspection reports, confirm that the
maxTokenparameter configuration covers the complete document content with no truncation errors.
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.