What this type of data looks like
Sources of TCM investment research data include official standards released by the National Pharmacopoeia Committee, provincial and municipal TCM processing specifications, public TCM component detection databases, clinical evidence-based literature, and origin traceability quality inspection reports. Update cycles: Official standards are revised centrally every 5 years, with supplementary announcements issued regularly. Component data is released irregularly as detection technology updates. Clinical literature is continuously added as research progresses. Each data document includes fields such as common name, Latin name, nature, taste and meridian tropism, functions and indications, usage and dosage, chemical components, processing methods, and precautions. Some documents include high-definition images and batch detection data. For field units: Dosage is mostly measured in grams and milliliters. Component content is measured in mg/g and %. Origin data includes longitude and latitude coordinates.
How these characteristics create constraints for database and operations
The multi-field and nested document structure requires the database to support flexible field expansion and full-text search. High-precision numerical values and batch information in component data require support for precise numerical queries and timestamp version management. Regular updates to official standards require operation processes to support batch data replacement and version rollback. Unstructured fragments of clinical literature require preprocessing before being uniformly stored in the vector database, while associating original literature metadata. Frequently updated component data requires configuring database read-write separation to reduce concurrent pressure.
How to set configurations
| Configuration Item | Recommended Value | Rationale |
|---|---|---|
PARSE_FILE_TIMEOUT_SECONDS | 1200 seconds | Single TCM documents may contain multiple image parsing tasks and long text fields, requiring extended parsing timeout |
UPLOAD_FILE_MAX_SIZE | 2000 MB | Supports uploading high-definition scans of batch detection reports and component data tables |
vectorStore.batchSize | 50 | TCM documents have a large number of fields. Control the number of entries per batch when inserting into the vector database to avoid timeouts |
Recall count | Top 15 entries | Covers multi-dimensional retrieval needs including nature, taste and meridian tropism, functions and indications, and components, increasing the recall base |
Similarity threshold | 0.75 | A precise threshold to distinguish similar TCMs, avoiding low-match irrelevant results |
db.versionRetentionCount | 10 | Retains historical versions of official standards to support compliance rollback and revision comparison |
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 specific analysis, and it is recommended to test on your own samples before finalizing.
Three common mistakes
- Symptom: The exported dataset.csv only contains the index field and no content field. Cause: The original text storage configuration of the vector database is not enabled, only index metadata is synchronized.
- Symptom: Request success rate is low in high-concurrency scenarios. Cause: Database read-write separation is not configured, and the vector database batch insertion threshold is set too high, leading to connection pool exhaustion.
- Symptom: Only part of the database table data is captured. Cause: The whitelist filter for table fields is not configured, the system skips unstructured remark fields, or the database connection timeout period is too short to complete full data pulling.
How to confirm the configuration is correct
- Upload a TCM standard document with multiple fields, check if all preset fields are included in the parsed database entries.
- Initiate a batch retrieval request, check the connection occupancy rate and request success rate on the system monitoring panel.
- Export the knowledge base dataset.csv, check if both index and content fields exist in the file.
- Configure a share link parameter, verify that the generated link carries the database identification id.
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.