What the data for this category looks like
Water utility marketing-related data sources include internal water user profile systems, pipe network operation and maintenance platforms, water conservation publicity material libraries, and regional water supply service bulletin boards of water utility enterprises. Update frequency follows these rules: User water usage ledger data is updated daily. Pipe network operation logs are updated monthly. Marketing materials are adjusted alongside activities or policies, with irregular frequency. Document structure includes structured water usage statistics tables and unstructured graphic and text publicity documents. Fields and units include standardized business fields such as "Water Supply Pressure (MPa)", "Average Daily Water Usage (Cubic Meters)", "Payment Overdue Days", as well as marketing management fields like material ID and release channel.
What constraints these characteristics impose on deployment and upgrade
Water utility data has many structured fields with specific units. Custom field mapping rules must be configured during deployment to prevent general parsing modules from misreading business parameters. Multi-format marketing materials include complex tables. Cross-format table extraction logic must be adapted when upgrading parsing modules to avoid loss of report content. Daily updated water usage data requires incremental synchronization. Scheduled pull tasks must be configured during deployment to avoid excessive system resource usage from full synchronization. Marketing material update frequency is irregular. Dynamic loading of the latest materials must be supported during upgrade. Hardcoding fixed paths is not recommended.
How to set the configurations
The following table lists configuration items, recommended values, and their rationales:
| Configuration Item | Recommended Value | Rationale |
|---|---|---|
PARSE_FILE_TIMEOUT_SECONDS | 300 seconds | Water utility marketing materials include water usage statistics tables with multiple rows and columns, which take longer to parse. The default 120 seconds is insufficient for long document parsing, and 300 seconds covers most scenarios. |
UPLOAD_FILE_MAX_SIZE | 2000 MB | Regional water usage statistics reports for water utilities often include multi-year data, with individual file sizes reaching 1.5 GB. Raising the upload limit prevents file truncation. |
maxContext | 8000–12000 characters | Water utility marketing content needs to be linked to user water usage profiles and business rules. A long context retains complete user information and marketing material details, adapting to multi-turn conversation scenarios. |
Recall count | Top 6 entries | Water utility marketing content covers three core scenarios: water conservation publicity, payment reminders, and pipe network maintenance. 6 recalled entries cover most user needs while avoiding information overload. |
Similarity threshold | 0.72–0.78 | The semantic similarity between user queries and marketing materials in water utility scenarios has small differences. A threshold that is too low introduces irrelevant content, while a threshold that is too high fails to match accurate marketing materials. |
Incremental sync interval | 1 hour | Daily updated water usage data needs to be synchronized to the knowledge base regularly. A 1-hour interval ensures the timeliness of marketing content while avoiding resource occupation from frequent requests to business systems. |
The parameter values provided on this page are common recommended starting points for configuration. Actual values are affected by material format, data volume, and business rules. Specific issues require individual analysis. It is recommended to test on samples specific to the deployment before finalizing values.
Three common mistakes
- Symptom: After deployment, calling the large model interface returns
400 Bad Requestwith the promptinvalid token encoder. Cause: The large model's API key and interface address are not configured correctly, or an incompatible model version is used, resulting in token parsing failure. - Symptom: The number of knowledge base recall results remains 0. Cause: The exclusive fields of water utility business data are not mapped correctly, or the similarity threshold is set too high, making it impossible to match user queries with marketing materials.
- Symptom: A large amount of table content is missing from parsed documents. Cause: Multi-format table parsing configuration is not enabled, and only plain text parsing mode is activated, making it impossible to extract structured table data from water utility reports.
How to confirm configurations are set correctly
- Upload a water utility marketing PDF containing regional water usage statistics tables. Check if the parsed result fully retains table structure and values, and adjust corresponding configuration items based on parsing performance.
- Simulate the user query "How to handle overdue water fees this month". Check if the recall results include corresponding payment reminder marketing materials, and adjust the similarity threshold and recall count based on recall results.
- View the running logs of incremental synchronization tasks. Confirm whether daily updated water usage data is synchronized to the knowledge base on time, and adjust the incremental synchronization interval based on synchronization delay.
- Test whether the generated marketing copy contains correct water utility business terms and units. Adjust the context window parameter based on generation results.
Question material comes from public community discussions. Configuration values are common starting points and should be measured against your own samples. Verified on 2026-09-14.