What the data for this category looks like
Joint-stock bank marketing content data primarily comes from internal marketing material libraries, campaign management systems, and historical customer inquiry records. Data update cadence aligns with marketing campaign cycles. Bulk updates are completed 1–3 days before a new campaign launches, and only minor adjustments to campaign validity periods and script details are made on an ongoing basis. Most documents are structured, including fields such as product_code, activity_name, target_crowd, activity_valid_period. Some promotional materials include unstructured text content. Time fields use the YYYY-MM-DD format, and text fields are counted in characters.
What constraints these characteristics impose on model access and configuration
The high proportion of structured data requires strict alignment of field mapping rules during model access, to avoid recalled marketing materials that do not match the target customer group due to field mismatches. Frequently updated marketing materials require a scheduled synchronization mechanism to ensure the latest campaign information is retrieved during model calls. Marketing content has strong timeliness, so the valid duration of the context window must be limited to avoid using outdated campaign data. Additionally, compliance requirements in financial scenarios require adding a sensitive information filtering step to model configuration, to prevent leakage of customer privacy and regulatory sensitive terms.
How to set configurations
| Configuration Item | Suggested Value | Rationale |
|---|---|---|
maxContext | 8000–12000 characters | Adapts to the total length of a single batch of marketing materials and historical conversation context, to avoid truncation of critical campaign information |
Recall count | Top 6–8 entries | Matches the number of materials per page in the marketing material library, to ensure recalled results cover all campaign dimensions |
rerankModelUrl | Locally deployed rerank service address | Meets compliance requirements for not allowing financial data to leave the region, to prevent external models from accessing internal marketing data |
syncSchedule | Every 4 hours | Aligns with the bulk update cadence of marketing materials, to ensure the latest campaign information is available during model calls |
sensitiveFilterEnable | Enabled | Filters customer privacy fields and regulatory sensitive terms in marketing materials, to meet financial compliance requirements |
toolCallEnable | Enabled | Supports calling internal tools based on marketing scenarios, such as querying campaign inventory and verifying campaign validity periods |
The parameter values provided on this page are all common recommended starting points for determining configurations. Actual values are affected by material form, data volume and business rules. Specific issues require case-by-case analysis, and it is recommended to test on your own samples before finalizing settings.
Three Common Mistakes
- Enabling tool calling without thought process output. Phenomenon: Model call results only return tool call instructions, with no intermediate thought steps. Cause: The
enableThoughtparameter is not enabled, or the model itself does not support tool call formats with thought processes. - Unable to select a model in the question classification configuration page. Phenomenon: No optional models appear in the dropdown menu, and a 400 status code is returned upon submission. Cause: The target model has not been added to the platform's available model whitelist, or the current FastGPT version is lower than v4.8.21.
- Abnormal recall result ranking after deploying the rerank model. Phenomenon: The returned material ranking does not match the configured
rerankReturnTopNparameter, or the rearranged results show no change. Cause: The interface address and authentication parameters of the rerank service have not been filled in correctly, or the input text length exceeds the model's supported limit.
How to Confirm Configuration Is Complete
- Run a model call test for a single marketing material, verify that the returned results include the latest campaign information, and that no field truncation or sensitive information leakage has occurred.
- Configure question classification rules, trigger a test call, confirm that the target model can be selected in the dropdown menu, and that no 400 status code errors are returned.
- Upload a batch of marketing materials, trigger a rerank rearrangement test, verify that the sorting logic of the returned results matches the configured requirements.
- View synchronization logs, confirm that the marketing material library is updated regularly according to the set
syncSchedule, with no synchronization failure records.
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.