Knowledge Base Retrieval and Recall for DTP Pharmacy

Data Characteristics in This Domain

DTP pharmacy pharmacovigilance data primarily originates from pharmaceutical manufacturers' periodic safety reports, daily pharmacy sales and patient medication feedback records, and national adverse drug reaction monitoring center announcements. This data updates frequently, especially when new drugs launch or during mass adverse events. Document structures typically include drug inserts, adverse reaction report forms, patient medication logs, and risk management plans. Beyond basic information like drug generic name, batch number, manufacturer, and dosage, fields also cover adverse event descriptions, occurrence time, severity, prognosis, intervention measures, patient underlying diseases, and concomitant medications. Units strictly adhere to medical and pharmaceutical standards, such as dosage units mg, g, IU, time units hours, days, weeks, and adverse reaction incidence percentages.

Constraints Imposed by These Characteristics on Knowledge Base Retrieval and Recall

The high update frequency of DTP pharmacy data requires the knowledge base to support rapid incremental indexing. This ensures the timeliness of retrieval results. The complex structure of drug inserts and adverse reaction reports, containing nested information and specialized terminology, challenges text segmentation and entity recognition. This necessitates more refined chunking strategies. Adverse event descriptions are often free text, including synonyms, abbreviations, and colloquialisms, which impacts retrieval accuracy. Recall must handle semantic similarity. Unstructured data in patient medication logs and feedback records, such as symptom descriptions, requires effective text preprocessing and standardization. Furthermore, precise matching of numerical fields like dosage and time requires support for multimodal or structured data joint queries during retrieval.

Configuration Recommendations

Configuration Item	Recommended Value	Rationale
`Chunk Size`	`800–1200 characters`	Drug inserts and adverse reaction reports often contain long professional descriptions. Maintaining a certain length helps preserve contextual completeness.
`Overlap Size`	`100 characters`	Ensures critical information is not split at chunk boundaries, improving recall accuracy.
`Recall Count`	`Top 10`	In pharmacovigilance, comprehensive judgment requires as much relevant information as possible to avoid omissions.
`Similarity Threshold`	`0.75–0.85`	Considering semantic differences in specialized terminology and free text, a relatively high threshold filters irrelevant results.
`UPLOAD_FILE_MAX_SIZE`	`500 MB`	Drug safety reports and risk management plans can be large files. Large file upload support is necessary.
`PARSE_FILE_TIMEOUT_SECONDS`	`600 seconds`	Parsing large PDF drug inserts or reports can take a long time.

Common Pitfalls

New drug inserts or adverse reaction reports uploaded after a knowledge base update are not retrievable. This occurs because the knowledge base re-indexing was not triggered or index building failed.
Retrieval results contain many document chunks irrelevant to the query intent. This happens when the chunking strategy is too coarse or the similarity threshold is set too low, leading to excessive noise in recall.
Querying the incidence rate of adverse reactions for a specific drug returns empty or inaccurate results. This occurs when the knowledge base fails to correctly identify and extract numerical fields and units from documents, or when the query does not account for numerical range matching.

Verification Steps

Upload a batch of documents containing the latest adverse reaction information. Immediately query to confirm new information is accurately recalled.
For a drug insert with various adverse reaction descriptions, retrieve using different keywords. Check if the returned document chunks are highly relevant to the query intent and cover key information in the insert.
Simulate a query for a dosage-related adverse reaction for a specific drug, such as "rash after 20mg of drug X". Check if the knowledge base identifies the 20mg dosage information and recalls corresponding reports.
Check the knowledge base backend index status and error logs. Confirm file upload and chunking processes are normal and index building is successful.

Note: The values provided are common starting points. Measure them against your 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.