tc_medical_base — 中醫典籍檢索語料

Clause-level (條文) corpus of five classical TCM texts with precomputed sentence-BERT embeddings, built for the 典籍提示 (classical-text hint) retrieval feature of the tcm-homevisit 中醫居家醫療 app (Taiwan).

Books (retrieval units after parsing)

書名	clauses
傷寒論（宋本）	603
傷寒雜病論（桂林古本）	995
金匱玉函經	1,215
金匱要略方論	590
傅青主女科	166
total	3,569

(古本康平傷寒論 was included in an earlier revision and removed 2026-08: its clauses overlap heavily with 宋本/桂林古本 with divergent wording, so the same 條文 crowded the top-k in three versions.)

Counts are retrieval units: one record per 條文/paragraph, with 方劑 (name + 組成 + 煎服法, the h-formula blocks) attached to the preceding clause's formula field rather than standing alone; front matter (序/凡例) excluded; paragraphs longer than 400 chars (傅青主女科) chunked at sentence boundaries. They therefore differ from the editions' nominal 條數.

Files

• corpus/tcm_classics.jsonl — one record per clause: {id, book, chapter, section, clause_no, text, text_embed, formula}. text keeps the original glyphs as published; text_embed has 異體字/古字 normalized (𤼵→發, 㣲→微, 隂→陰, …) and whitespace stripped — embed and search against text_embed, display text.
• corpus/embeddings.npy — float32 (3569, 512), L2-normalized, row i ↔ jsonl line i. Model: BAAI/bge-small-zh-v1.5, passages encoded without instruction prefix. Queries must be encoded with the prefix 為這個句子生成表示以用於檢索相關文章： then scored by dot product.
• raw/*.html — cached source HTML for reproducibility.
• scripts/build_corpus.py, scripts/build_embeddings.py — full rebuild:

pip install -r requirements.txt
python scripts/build_corpus.py      # fetch + parse → corpus/tcm_classics.jsonl
python scripts/build_embeddings.py  # encode → corpus/embeddings.npy

Source & attribution

Text source: 中醫笈成 (jicheng.tw), data repository gitlab.com/jicheng/jc.data. The classical texts themselves are in the public domain; the punctuated and collated digital editions are the work of 中醫笈成 — please keep this attribution when reusing.

Intended use

Advisory retrieval hints (僅供參考) for licensed TCM practitioners; not a diagnostic or prescribing system.