Data Characteristics for this Category
Patient Assistance Program (PAP) data originates primarily from official documents published by pharmaceutical companies, foundations, and medical institutions. These documents are typically in PDF, Word, or structured data formats. They contain patient application criteria, drug lists, assistance processes, review standards, and contact information. Document update frequencies vary; some programs may revise annually, while others update intermittently based on policy changes. Document structures usually include chapter headings, clause numbers, tables, and attachments. Common fields include generic drug names, brand names, indications, patient disease types, income verification requirements, and donated drug quantities. Some fields may contain specific medical or legal terminology. Units involve monetary amounts, quantities, and time periods.
Constraints from these Characteristics on "Citation and Traceability"
The official and rigorous nature of PAP documents requires precise citation in the Q&A system to ensure authoritative and credible answers. The uncertain update frequency demands a knowledge base that can flexibly handle document version iterations and accurately point to the currently effective program version during Q&A. Tables and specific fields within documents, such as drug lists or income standards, impose requirements on text segmentation and retrieval strategies. This avoids critical information loss or incomplete contextual semantics due to improper segmentation. Citation sources must be precise down to the page number, chapter, or paragraph of the original text. This traceability capability is crucial, especially when patient rights and compliance issues are involved.
Configuration Settings
| Configuration Item | Recommended Value | Rationale for this Value |
|---|---|---|
Chunk size (Segment Length) | 800–1200 characters (characters) | Balances the chapter logic and semantic completeness of PAP documents. Avoids cutting context too short or introducing irrelevant information when too long. |
Chunk overlap (Segment Overlap) | 200 characters (characters) | Ensures sufficient overlap between paragraphs to handle cases where critical information spans across segments, improving recall rate. |
Recall count (Recall Count) | Top 5 entries (top 5) | PAP Q&A demands high accuracy. Increasing the recall count appropriately raises the probability of selecting highly relevant original text snippets. |
Similarity threshold (Similarity Threshold) | 0.75 | Ensures recalled text snippets are highly relevant to the user's question, reducing citations of inaccurate or low-relevance content. |
Rerank result count (Reranked Return Count) | 3 entries (3) | After reranking, focuses on the 3 most relevant original text snippets, improving answer precision and traceability efficiency. |
Citation Metadata | Page number, Chapter title, Document version number | PAP documents have high traceability requirements. Providing detailed metadata helps users quickly locate the original source. |
Three Common Mistakes
- Q&A results do not reflect local knowledge base content, but the citation list shows relevant documents. This may occur because the large language model failed to fully utilize the recalled contextual information during the content generation phase of the response.
- API calls return citation information missing page numbers or chapters. This may be due to incorrect identification or storage of this information during document parsing or metadata extraction when building the knowledge base.
- The system provides inaccurate answers to questions about specific drug names or assistance conditions. This may be due to a document segmentation strategy that failed to effectively process critical information within tables or nested structures, leading to incomplete recalled snippets.
How to Verify Configuration
- Ask questions about key PAP clauses. Verify that the cited original text in the answer points to the correct document, page, and chapter.
- For critical information in documents containing tables or lists (e.g., drug lists, application materials), check if the Q&A system can accurately extract and cite the relevant rows or cells.
- Simulate user questions about updates or changes to a specific PAP. Verify that the system can cite the latest version of the document and highlight version differences.
The values provided are common starting points and should be measured against the reader's own 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.