Data Characteristics
Medical insurance settlement quality documents originate from policy regulations, operational guidelines, internal settlement procedures, audit cases, and FAQs. These are issued by various levels of medical insurance bureaus. Documents update frequently, sometimes monthly or weekly, especially with policy adjustments and annual settlement rule changes. Document structures vary: PDF policy texts, Word internal guidelines, Excel cost examples, and online technical specifications. Fields and units are highly specialized, including "medical insurance payment scope," "payment ratio," "self-payment amount," "overall fund payment," "deductible," and "cap." They often involve disease codes (ICD-10), project codes (C-DRG/DIP), generic drug names, and consumable batch numbers. Regional medical insurance policies can lead to different value logic for the same fields.
Constraints on Knowledge Base Retrieval and Recall
Frequent medical insurance policy updates require the knowledge base to support rapid updates and version management, ensuring timely retrieval results. Diverse document formats necessitate robust file parsing capabilities, especially for accurate extraction of tabular data and unstructured text. Highly specialized fields and units demand precise entity recognition and semantic understanding to avoid errors from ambiguous terminology. Regional differences mean the knowledge base needs to support multi-version or multi-region data isolation and retrieval to prevent policy confusion. Additionally, the complexity of medical insurance settlement issues often involves cross-referencing multiple policies. This challenges the retrieval system's multi-hop question answering and related information recall capabilities, requiring precise identification of key supporting information from extensive details.
Configuration Settings
| Configuration Item | Recommended Value | Rationale |
|---|---|---|
Chunk size (Segment Length) | 800–1200 characters | Medical insurance policy clauses are often long with strong contextual relevance. Increasing segment length retains more contextual information. |
Chunk Overlap Length (Segment Overlap Length) | 100 characters | Ensures sufficient overlap between adjacent segments to handle retrieval boundary issues and maintain context continuity. |
Recall count (Recall Count) | Top 5–8 entries | Medical insurance questions often require multiple perspectives and policy support. Increasing recall count improves coverage. |
Similarity threshold (Similarity Threshold) | 0.78–0.85 | Medical insurance terminology is highly specialized, requiring high precision for recall. This balances relevance with avoiding excessive recall. |
Rerank result count (Reranked Return Count) | 3 entries | After reranking by the large model, select the most relevant few entries to improve final output quality. |
PARSE_FILE_TIMEOUT_SECONDS | 600 seconds | Parsing large files (e.g., long policy PDF documents) can be time-consuming. A longer timeout prevents parsing failures. |
Common Pitfalls
upstream connect error or disconnect/reset before headwhen uploading documents: This usually occurs because the uploaded file size exceeds the default limit of the reverse proxy (e.g., Nginx), or file parsing takes too long, causing the backend service connection to terminate.- After importing an Excel file into the knowledge base, a single policy is split into multiple rows, or multiple policies are merged into one row: This happens when the custom delimiter is not configured correctly, preventing the system from identifying valid row boundaries in the Excel file.
- Retrieval results contain policy clauses irrelevant to the query: This may result from a
Similarity threshold(Similarity Threshold) set too low, leading to the recall of semantically unrelated document fragments.
How to Verify Configuration
- Upload typical medical insurance settlement documents in various formats (PDF, Word, Excel). Check if files are parsed successfully and knowledge segments are generated.
- Perform a search for specific medical insurance settlement issues, such as "DRG payment scope." Observe if the recalled document fragments accurately contain relevant policy clauses and key fields.
- Use complex queries containing specific medical insurance terminology. Check if the results in
Rerank result count(Reranked Return Count) directly answer the question, and compare them with un-reranked results. - Simulate a medical insurance policy update scenario by uploading a new version of a policy file. Verify the timeliness of retrieval results after the knowledge base update.
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.