Data Profile for This Category
Cosmetics marketing data for customer acquisition scenarios comes from four main sources: brand official product libraries, compliance filing platforms, e-commerce platform user reviews, and review materials from beauty content platforms. Brand product libraries contain structured fields such as ingredient proportions, applicable skin types, specification units, and filing numbers. Compliance filing platforms provide official basis for efficacy claims. Data is updated in batches when new products launch. Daily updates only make minor tweaks to individual product parameters. Unstructured data includes past marketing copy and live script snippets. It is stored as plain text or rich text, with some entries linked to structured product information.
Constraints for Multi-turn Dialogue and Prompt Engineering
Cosmetics marketing dialogue for customer acquisition must balance compliance requirements and user personalized needs. Compliance rules for cosmetics data require multi-turn dialogue to verify generated content for efficacy claims, to avoid non-compliant statements. The precision requirements for structured fields mean prompts must explicitly specify the field format for calling product libraries. This prevents generation of inaccurate product information. Multi-turn dialogue must repeatedly confirm user details such as skin type and usage scenarios. Recommended content varies widely based on different adaptation conditions. Unstructured marketing materials must use user historical dialogue context to avoid duplicate recommendations or disconnected content. Context length must also be controlled to stay within model processing limits.
Configuration Settings
| Configuration Item | Recommended Value | Rationale |
|---|---|---|
maxContext | 8000–12000 characters | Cosmetics marketing conversations need to retain multi-turn context including user skin type, usage scenarios, past recommendations, and similar details, to avoid losing key information |
systemPromptTemplate | Embed compliance verification rules + product structured field calling template | Cosmetics marketing content must comply with advertising compliance requirements, and must accurately call structured data such as filed ingredients and efficacy claims |
recallTopK | Top 6–8 entries | There are many marketing materials and product information related to cosmetics; too many recalled entries will lead to redundant context, while too few will miss key adaptation information |
codeRunTimeout | 60 seconds | Compliance verification interfaces or product library queries need to be called, and sufficient time must be reserved for scenarios with long processing times |
customSessionIdEnable | Enabled | Session history must be associated with the user's customUid to avoid confusion across user sessions |
removeThinkTag | Enabled | Generated marketing content must remove thought processes to avoid exposing internal logic, meeting output requirements |
The parameter values provided on this page are common starting points for configuration. Actual values are affected by material form, data volume, and business rules. Specific issues require individual analysis, and it is recommended to test on your own samples before finalizing settings.
Three Common Misconfiguration Issues
- Symptom: After calling the API with a specified customUid, the retrieved session records include content from other users. Cause: The
customSessionIdEnableconfiguration is not enabled, and the association between sessions and customUid is not established. - Symptom: Think tags remain in generated marketing content. The configuration works during testing but fails to take effect in the production environment. Cause: The
removeThinkTagparameter is not properly configured in the workflow, or the parameter is not synchronized to the production deployment environment. - Symptom: Unauthorized efficacy claims appear in generated marketing content. Cause: The prompt does not enforce calling official product library data, and only relies on general knowledge for content generation.
How to Verify Correct Configuration
- Initiate a test session with a specified customUid, call the session history interface, and confirm that the returned sessions only include dialogue content for that customUid.
- Run a workflow that includes the
removeThinkTagconfiguration, view the final output content, and confirm that there is no content wrapped in think tags. - Input test content including efficacy queries into the dialogue node, and confirm that the generated marketing content does not include unauthorized efficacy claims.
- Adjust the
recallTopKparameter, and check that the number of recalled product materials matches the preset value range.
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.