concept-graph / README.md
Rahvin5677's picture
concept-graph: sanitized card + SPEC.md (flip prep)
2396bc1 verified
|
Raw History Blame Contribute Delete
9.78 kB
---
license: cc-by-4.0
status: active
language:
- mul
task_categories:
- feature-extraction
tags:
- nuberea
- concept-graph
- thematic-analysis
- old-testament
- new-testament
- biblical-studies
size_categories:
- 10K<n<100K
extra_gated_prompt: >-
This dataset is distributed under CC BY 4.0. Access is gated for provenance
and attribution stewardship; please review the attribution note in the
dataset card.
configs:
- config_name: span_node
data_files:
- split: train
path: data/span_node/train-*.parquet
- config_name: span_edge
data_files:
- split: train
path: data/span_edge/train-*.parquet
- config_name: span_signature
data_files:
- split: train
path: data/span_signature/train-*.parquet
- config_name: span_feature_edge
data_files:
- split: train
path: data/span_feature_edge/train-*.parquet
- config_name: feature_anchor
data_files:
- split: train
path: data/feature_anchor/train-*.parquet
- config_name: community_candidate
data_files:
- split: train
path: data/community_candidate/train-*.parquet
- config_name: community_membership
data_files:
- split: train
path: data/community_membership/train-*.parquet
- config_name: concept_node
data_files:
- split: train
path: data/concept_node/train-*.parquet
- config_name: concept_span_edge
data_files:
- split: train
path: data/concept_span_edge/train-*.parquet
- config_name: nt_concept_centroid_reliability
data_files:
- split: train
path: data/nt_concept_centroid_reliability/train-*.parquet
- config_name: concept_concept_edge
data_files:
- split: train
path: data/concept_concept_edge/train-*.parquet
- config_name: centroid_index_l1
data_files:
- split: train
path: data/centroid_index_l1/train-*.parquet
- config_name: centroid_index_l2
data_files:
- split: train
path: data/centroid_index_l2/train-*.parquet
- config_name: inheritance_field_nodes
data_files:
- split: train
path: data/inheritance_field_nodes/train-*.parquet
- config_name: inheritance_field_edges
data_files:
- split: train
path: data/inheritance_field_edges/train-*.parquet
---
# NuBerea Concept Graph
A cross-testament concept graph over the canonical Bible (Old and New Testament), plus a
second, separate reception/discourse graph unifying edge endpoints across seven reception
surfaces. Includes thematic span nodes, inter-span edges, lexical/syntactic/source-critical
feature anchors, graph-community detection results, and a BERTopic-derived concept layer
(concept nodes, concept-to-concept edges, concept-to-pericope edges).
## Dataset description
- 15 loadable configs, ~2.31M rows total (dominated by the two inheritance-field tables; the
concept/span/community family below that is 10K-100K rows)
- License: CC BY 4.0
- No source-text prose is shipped in any config: every table carries ids, structural/lexical
labels, computed weights or scores, and short bibliographic pointers only.
## Configs
| Config | Rows | Cols | Grain | Description |
|---|---|---|---|---|
| `span_node` | 33,575 | 12 | verse + pericope | Thematic span registry: corpus, span grain (`verse`\|`pericope`), OSIS start/end, canonical book. `parent_span_id` links a verse row to its containing pericope. |
| `span_edge` | 22,551 | 12 | span-to-span | Edges between thematic spans: edge family (structural / lexical-syntactic-similarity / source-critical / bridge-crosslink), type, weight, confidence class, directionality. |
| `span_signature` | 33,575 | 10 | verse + pericope | Per-span multi-dimensional feature signature: lemma / morphology / syntax / source-critical / bridge / crosslink profile, each a JSON-serialized object; quality flags. |
| `span_feature_edge` | 8,770 | 6 | span-to-anchor | Links a span to a contributing feature anchor, with salience and evidence family. |
| `feature_anchor` | 142 | 7 | registry | Named lexical, syntactic, and source-critical feature anchors with display name, corpus scope, and span count. |
| `community_candidate` | 48 | 7 | community-level | Graph-clustering community summaries: dominant corpus/anchor, book coverage, size. |
| `community_membership` | 2,423 | 6 | pericope | Pericope-to-community membership (span-community pairs). |
| `concept_node` | 86 | 18 | concept registry | BERTopic-derived concepts: 78 L1 clusters + 8 L2 super-concepts, with a 384-dim centroid vector, top terms, span counts, corpus scope. |
| `concept_span_edge` | 2,412 | 10 | pericope | Concept-to-pericope assignment edges, canonical Bible only, with soft-probability weight and relation type (primary/secondary/allusive). |
| `nt_concept_centroid_reliability` | 86 | 12 | concept-level | Per-concept NT-vs-OT activation reliability diagnostics (span counts, OT-pull dominance, a frozen reliability score/flag). |
| `concept_concept_edge` | 298 | 11 | concept-to-concept | Concept relatedness edges (L1-L1, centroid similarity) and hierarchy edges (L1-to-L2). No verse/span anchor — a pure id-to-id scaffold. |
| `centroid_index_l1` | 78 | 3 | index | Row-index lookup into the (non-tabular) L1 centroid matrix. |
| `centroid_index_l2` | 8 | 3 | index | Row-index lookup into the (non-tabular) L2 centroid matrix. |
| `inheritance_field_nodes` | 342,380 | 8 | node graph, no verse anchor | Every reception/discourse edge endpoint across 7 corpus surfaces, unified to one `node_id`, with an authored-or-received time signal (~95% time-placeable). |
| `inheritance_field_edges` | 1,865,640 | 6 | edge graph, no verse anchor | Reception-plus-discourse edges re-keyed onto `node_id`, tagged `inter_field_transmission` or `intra_field_discourse`. |
### Internal artifacts (not loadable configs)
The following directory under `data/` is a build-time intermediate kept alongside the dataset
for provenance and reproducibility. It is **not** registered as a `load_dataset` config.
- **`data/centroids/`** — raw concept-centroid embeddings and their index sidecars (`.npy` float
arrays plus JSON index maps). Not tabular parquet; mirrors information already exposed through
`concept_node`'s `centroid` column and the two `centroid_index_l*` tables above.
This artifact may change or be removed between builds without notice; treat the 15 registered
configs above as the stable public surface.
## Two graphs, one repo
This repo carries two structurally distinct graphs that share no columns:
1. **The span/concept graph** (`span_*`, `feature_anchor`, `community_*`, `concept_*`,
`nt_concept_centroid_reliability`, `centroid_index_l*`) — thematic spans over the canonical
Bible (OT+NT), their features, and the BERTopic concept layer built on top of them.
2. **The inheritance field** (`inheritance_field_nodes`, `inheritance_field_edges`) — a separate
reception/discourse transmission graph spanning seven corpus surfaces beyond the canonical
Bible (patristic writing, later reception corpora, New Testament, Old Testament, deuterocanon,
classical Greek, and Second-Temple-era Jewish writing). Its nodes are keyed by a content hash
of the writing's own native citation, not by OSIS reference, and most rows carry no Bible-verse
anchor at all.
## Granularity
`span_node` and `span_signature` keep both grains side by side: `span_grain` distinguishes a
verse row from a pericope row, and every pericope row's constituent verses remain separately
present as their own verse-grain rows in the same table (verse rows are never dropped in favor
of the pericope rollup). `concept_span_edge` is pericope-grain only in this repo; a
verse-grain concept-assignment layer for the canonical Bible, plus ten additional reception
corpora, is published as a separate config in the companion `concept-analysis` repo (see below).
The `inheritance_field_*` pair has no per-row Bible-verse anchor at all — most of the underlying
writings (patristic and later reception literature, Second-Temple-era Jewish writing) are never
verse-anchored to begin with. This is disclosed, not implied to be otherwise: treat these two
tables as a distinct, node-graph-only surface, not a rollup over the verse-anchored tables above.
## Companion repo
`concept_node`, `concept_span_edge`, `concept_concept_edge`, and `community_membership`/
`span_node` in this repo are the concept-graph's own registry and canonical-Bible edges. A
sibling repo, `NuBerea/concept-analysis`, publishes the corresponding **verse- and segment-grain**
concept-assignment tables — one per corpus, covering the canonical Bible plus ten additional
reception corpora (patristic Greek writing, Apostolic Fathers, Deuterocanon, early Christian
writing, Josephus, Nag Hammadi, Philo, Pseudepigrapha, Talmud, and Targum) — along with
concept-drift and concept-trajectory tables built over the same concept ids used here.
`concept-analysis`'s `community_membership` config is an import of this repo's own
`community_membership` table, republished there for that repo's own consumers.
## What this is not
This is a machine-computed thematic/topological layer, not a scholarly commentary and not
ground truth. Community and concept assignments are the output of unsupervised clustering
(BERTopic, weighted label propagation) over aligned lexical/syntactic/source-critical features,
not adjudicated theological judgments. No source-text edition is redistributed here; every
config ships ids, structural/lexical labels, computed weights or scores, and short bibliographic
pointers (OSIS references, CTS-style content identifiers) only.
## Attribution
NuBerea project. Licensed CC BY 4.0. Built over the estate's aligned Hebrew and Greek
morphology/syntax layers and source-critical span registry for the canonical Bible; specific
constituent-edition citations for those substrate layers are not itemized in this card.