emadjumaah/hadith-kg

الجامع — Hadith Knowledge Graph (SQLite)

A knowledge graph of the sciences of hadith (علوم الحديث), offered openly as a donation for students and scholars:

• 715,790 hadith from 425 classical collections
• 577,024 complete isnad chains as ordered narrator links (4.4M edges)
• 49,819 narrator biographies — tabaqa, reliability grading, tadlis/ikhtilat flags, dates and places
• 127,863 jarh-wa-ta'dil statements by 1,015 critics, fully entity-linked
• 20,745 meaning groups clustering every narration of one matn across books, with 768-dim Gemini embeddings for semantic search
• 54,862-node subject tree linked down to the hadith level
• Hadith text stored as plain Arabic with character-offset annotations for narrator mentions, matn boundaries, and Quran quotes

Files

File	Size	What it is
hadith-kg.db	~1.6 GB	Canonical relational knowledge graph (SQLite) — normalized tables, chains, criticism, topics, embeddings
hadith-app.db	~2.9 GB	Application database (monlite) — document collections + Arabic FTS, precomputed dashboards

Open with any SQLite client. Table/semantics documentation lives in the meta table of each database and in the project README.

Application

The full open-source platform (API server, RTL dashboard with isnad-tree visualizations, semantic search, cited RAG chat): https://github.com/emadjumaah/hadith

Notes for researchers

• sanad_rawis.pos = 0 is the collection author; the chain ascends to the companion; transmission direction is Prophet ﷺ → companion → … → author.
• sanads.length counts distinct narrators; sanads.max_rank includes the author. meaning_groups.sahaba_qty/repeat_qty are legacy declared values — compute live statistics from the chains for exact numbers.
• A small number of chain narrators are stubs (rawis.is_stub = 1) whose biographies were absent in the source; names were recovered from the hadith text itself.