Data Characteristics
CAR-T cell therapy regulations and Standard Operating Procedure (SOP) documents originate from guidelines published by regulatory bodies such as the National Medical Products Administration (NMPA), the U.S. Food and Drug Administration (FDA), and the European Medicines Agency (EMA). They also include quality management system documents developed by individual medical institutions. These documents update infrequently, typically every few months to several years, in response to regulatory policy changes or technological advancements. Document structures are rigorous, often in PDF or Word format, containing extensive professional terminology, flowcharts, tables, and legal citations. Fields cover cell preparation, patient management, quality control, and adverse event handling. Units often follow international standards (e.g., volume in mL, cell count in 10^6 cells/kg), with precise requirements for parameters like temperature and time.
Constraints on "Citation and Traceability"
The authoritative and rigorous nature of CAR-T cell therapy regulatory documents demands precise and error-free source citations. This ensures the legal basis of answers is verifiable. Low update frequency highlights the importance of historical version management, requiring the ability to trace documents to specific points in time. The prevalence of specialized terminology, including synonyms or near-synonyms, increases the demands on semantic matching and recall accuracy. Tables and flowcharts mean traditional text segmentation may not capture complete information effectively, necessitating smarter parsing strategies. Furthermore, citing precise parameters like temperature and time requires the system to accurately identify and present these values, avoiding ambiguity and ensuring the practicality and safety of responses.
Configuration Strategy
| Configuration Item | Recommended Value | Rationale |
|---|---|---|
Chunk size | 500–800 characters | Balances the completeness of long paragraphs with the precision of short ones, adapting to regulatory document structures. |
Recall count | Top 8–12 entries | Increases the probability of recalling relevant segments, covering multiple aspects of information for complex queries. |
Similarity threshold | 0.75–0.85 | Ensures high relevance while allowing for variations in specialized terminology. |
Rerank result count | Top 5 entries | Selects the most relevant and authoritative citation snippets, reducing redundant information. |
citation Link Display type | document name + page number | Allows users to quickly locate specific positions in original documents, meeting traceability requirements. |
Knowledge Base Version Control | Enabled | Ensures answers are based on specific time-point versions of regulations, supporting historical traceability. |
Common Pitfalls
- Missing citation filenames or returning
quote type error: This often results from incorrect parsing of document metadata in the knowledge base configuration, such as failing to extract filenames or page numbers during document upload. - Quoted numerical values in answers do not match the original text: This may occur if the document parser fails to accurately identify numbers with units, or if the segmentation strategy separates critical numerical values from their context.
- Inability to cite key information from tables or flowcharts: Traditional text segmenters offer insufficient support for non-textual content, preventing structured or semi-structured data from being effectively indexed and recalled.
Verification Steps
- Test with typical CAR-T cell therapy regulatory questions. Check if the document names and page numbers cited in the answers are consistent and accurate with the original documents.
- Validate queries involving specific parameters (e.g., temperature, time, dosage). Ensure the numerical values cited in the answers precisely match the original text and that units are correct.
- Randomly select documents containing tables or flowcharts. Ask relevant questions and verify if the system can correctly cite this information or indicate its location.
- Test with updated regulatory documents. Confirm the system identifies and prioritizes the latest version while still allowing traceability to older versions through specific query methods.
The values provided are common starting points and should be measured against specific samples.
Question material comes from public community discussions. Configuration values are common starting points and should be measured against your own samples. Verified on 2026-09-21.