What the data for this category looks like
Cosmetics research report data mainly comes from public analysis from industry consulting firms, public product filing information from brands, public reports from compliance testing institutions, and aggregated consumer feedback from e-commerce platforms. Update schedules vary by content type: brand filing information updates in real time with new product launches, industry research reports are released quarterly, and test reports are updated with inspection batches.
Single documents typically include product basic information, compliance filing number, efficacy claim basis, ingredient details, and market performance modules. Some documents include competitor comparison tables. Fields include product name, filing number, test item, test result, issuing organization, and release date. No unified fixed statistical units are used, and some fields adjust based on test items.
What Constraints These Characteristics Impose on the Citation Source and Traceability Link
As research report sources are scattered and include different types of content such as filings, tests, and industry analyses, the traceability link must distinguish exclusive identifiers for different sources. For example, filing information must be linked to drug regulatory filing numbers, and test reports must label testing institutions and items.
Ingredient details are tightly bound to compliance fields. Citations must link corresponding test results and filing information at the same time to avoid traceability failure from citing efficacy descriptions alone. Comparison tables and multi-module content in documents require retaining complete contextual associations during parsing. Split citations will lose key comparison information otherwise.
Differences in update frequencies across content also require the traceability link to filter invalid expired content by time range.
How to Configure
| Configuration Item | Recommended Value | Rationale |
|---|---|---|
Recall Count | Top 8-12 entries | Cosmetics research reports cover multi-dimensional modules including ingredients, efficacy, and compliance. A sufficient number of recalled entries is needed to cover core information and avoid missing key content |
Similarity Threshold | 0.75-0.85 | Cosmetics research reports have relatively professional content. This range filters low-match irrelevant content while retaining weakly matched entries related to specific efficacy claims |
Citation Template | 【Source: {source}, Issuing Organization: {org}, Release Date: {date}, Filing Number: {record_no}】 | Cosmetics research report traceability requires clear labeling of exclusive fields such as filing numbers and issuing organizations to ensure users can trace compliance and product information |
Segment Length | 800-1200 characters | Cosmetics research reports include ingredient details and comparison tables. Longer segments can retain complete module information and avoid breaking content relevance by splitting |
Knowledge Base Time Range Filter | Only recall documents from the past 12 months | The cosmetics industry has fast product iteration. Expired research reports have low reference value, and filtering improves traceability accuracy |
Reranked Return Count | Top 5 entries | Prioritize displaying the most relevant core research report content to avoid excessive redundant information interfering with traceability judgments |
The parameter values provided on this page are common recommended starting points for configuration. Actual values are affected by material form, data volume and business rules. Specific issues require case-by-case analysis, and testing should be performed on local samples before finalizing values.
Three Common Misconfigurations
- Phenomenon: The workflow's returned answer does not fully reuse the original answer from the question-and-answer pairs in the knowledge base, and includes additional generated content. Reason: The
Force Reference Knowledge Baseparameter is not enabled, or the prompt does not explicitly require answering only based on knowledge base content. - Phenomenon: The citation module of the returned result is empty or does not display source information. Reason: The
Citation Templateparameter is not configured, or the template does not include necessary metadata fields such as {source} and {record_no}. - Phenomenon: The variable selection panel of the code running node cannot find the entry fields returned by the knowledge base search. Reason: The
Return Original Document Fieldsswitch is not enabled in the knowledge base search node, causing document metadata to not be exposed.
How to Verify Proper Configuration
- Launch a test query, enter a specific cosmetics-related question, and check that the answer fully matches the original text in the knowledge base with no additional generated statements.
- Check the citation module at the bottom of the answer to confirm that it includes exclusive fields such as issuing organization, filing number, and release date, with a format consistent with the configured
Citation Template. - Check the number of returned citation entries to confirm that it matches the configured value of
Recall Count, with neither too many nor too few recall results. - Check the system logs to confirm that all recalled documents meet the configuration requirements of
Knowledge Base Time Range Filter, with no expired documents mixed in.
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.