Data Profile for This Category
Data for this category primarily comes from the 12345 citizen hotline platform, contractor business systems, knowledge bases, follow-up evaluation data, and local hotline management regulations. The update frequency is daily.
Commonly processed documents include request registration forms, dispatch records, handling responses, follow-up records, extension applications, return explanations, and knowledge base entries. Typical fields include request number, responsible entity, request content, response guidelines, follow-up results, and others. There is no unified fixed format. Some data is structured data exported from business systems, while some is unstructured text entered manually.
Constraints Imposed on Vector Models and Indexing
Data sources are scattered and have diverse formats, including both structured exported business system data and unstructured manually entered text. This requires index configurations to support vector extraction for multiple data types.
Daily updated work order data fluctuates with business volume. Incremental indexing must be configured to avoid resource consumption from full reindexing.
Documents cover the entire work order process. Vector recall must accurately match key information across different stages. For example, request type judgment must link to registration form content. Response generation must reference handling records and knowledge base entries.
Additionally, work order processing has timeliness requirements. Vector retrieval response speed must fit the time window of the business process, to avoid work order overdue due to retrieval delays.
Configuration Settings
| Configuration Item | Recommended Value | Rationale |
|---|---|---|
| Chunk Length | 800–1200 characters | Fits the typical length of 12345 hotline work order documents, avoids losing semantic connections due to overly large single chunks |
| Overlap Length | 100–200 characters | Connects context between adjacent chunks, ensures complete semantics of requests and responses are covered by vectors |
| Similarity Threshold | 0.75–0.85 | Calibrated based on actual testing, balances accuracy and recall for request type matching, avoids misjudgment or missed judgments |
| Single Search Limit | Top 5 entries | Focuses on the most relevant work order history and knowledge base content, reduces interference from irrelevant information on response generation |
maxToken | 4096 | Fits the context length limits of general large models, ensures recalled document content can fully fit into the generation window |
defaultToken | 1024 | Balances content completeness and response speed for response generation, fits the typical length of work order responses |
Three Common Errors
- Enabling the Doubao-embedding-large indexing model, filling in a custom request address and apikey, then testing directly results in an error. The cause is failing to confirm network connectivity of the custom request address, or the apikey not being authorized to access the vector model service.
- Refreshing the knowledge base page displays "No available indexing models detected". The cause is failing to associate the configured vector model with the current knowledge base, or the model service health check failing.
- Enabling multi-vector support for table data, but recall results do not match table field information. The cause is failing to configure dedicated vector extraction rules for structured fields of work order-style tables, leading to incorrect vectorization of table content.
How to Confirm Proper Configuration
- Navigate to the knowledge base settings page of FastGPT v4.17.0, check the vector model configuration items, confirm the status of the enabled model shows normal.
- Upload a typical request registration form document, perform a vector recall test, check that the number of returned matching entries falls within the expected range.
- Create a new conversation agent, enter a query related to work orders, confirm that the index can be normally called to obtain context content, with no error prompts.
- Check the monitoring logs of the indexing service, confirm there are no records of connection timeouts, parameter errors, or permission exceptions.
The values provided are common starting points, and should be evaluated against the user'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.