---
license: cc-by-sa-4.0
task_categories:
- graph-ml
- tabular-classification
language:
- en
tags:
- wikipedia
- topology
- edge-list
- networkx
- bipartite-graph
- structural-alignment
- casei
pretty_name: WikiLoL (Wiki Lists-of-lists)
size_categories:
- 10M<n<100M
---

# WikiLoL-Epoch-20260503: Wiki Lists-of-lists

## IMPORTANT 

There exist negative results in one or more of the derivative analyses. Manual intervention is required to avoid poisoning of downstream tasks, until a repair and update can be finalized. The main data is intact. 

## Dataset Description
**WikiLoL** is a deterministic, structural edge-list extraction of the English Wikipedia's list ecosystem. 

Rather than extracting narrative wikitext, this dataset isolates the explicit routing mechanisms humans use to hierarchically categorize concepts. It extracts the underlying Abstract Syntax Trees (ASTs) of Wikipedia's "Lists of lists", standard wikitables, and bulleted indices. This compiles into a massive bipartite graph connecting curated index pages to their target entities, ultimately projected into a unipartite semantic co-occurrence matrix.

**Temporal Drift & Reproducibility:**
This dataset represents a static extraction of the English Wikipedia dump acquired on **May 3, 2026**. Wikipedia's internal topology is highly volatile. WikiLoL is intentionally frozen at this epoch to provide a mathematically stable baseline for reproducible research. Engineers embedding this graph into vector space or training Graph Neural Networks can be assured that the adjacency matrices will not shift during longitudinal studies.

## Repository Artifacts

### 1. The Core Topologies
* `wikilol_dataset.parquet` (1.95 GB): The raw, un-nested AST JSON extraction of all 311,862 source lists.
* `wikilol_edgelist_canonical.parquet` (237.39 MB): The unweighted 15.4-million-edge `[Source -> Target]` baseline bipartite map.
* `wikilol_edgelist_taxonomic.parquet` (238.35 MB): The enriched edge list, featuring boolean routing flags (`target_is_list`, `target_is_internal`) to trace the 1-degree list ecosystem.

### 2. The Node Feature Matrices
* `wikilol_node_features.parquet` (137.12 MB): A computed physics table containing In-Degree, Out-Degree, Global PageRank, HITS decomposition (Hubs/Authorities), and Cross-Domain Bridge Scores (Shannon entropy) for all 6.3 million targets.
* `wikilol_adjacency.npz` (31.71 MB) & `wikilol_node_index.json` (226.79 MB): The 15.4M edge Bipartite CSR matrix and its integer-to-text mapping.
* `wikilol_cooccurrence.parquet` (11.98 GB) & `wikilol_entity_index.json` (220.30 MB): A 4.3-billion-edge `[Node_A, Node_B, Weight]` upper-triangle unipartite mesh calculating the shared-list proximity of target entities.
* `wikilol_list_similarity.parquet` (253.42 MB) & `wikilol_list_index.json` (6.10 MB): A unipartite matrix calculating the shared-entity overlap between the list pages themselves.

### 3. Domain Subgraphs
Pre-partitioned topologies based on global category distributions, allowing for isolated network analysis:
* `subgraphs/wikilol_subgraph_stem.parquet`
* `subgraphs/wikilol_subgraph_history.parquet`
* `subgraphs/wikilol_subgraph_geography.parquet`
* `subgraphs/wikilol_subgraph_popculture.parquet`
* `subgraphs/wikilol_subgraph_sports.parquet`

### 4. Audit & Schema Logs
* `wikilol_macrotopology.json` (< 1 MB)
* `wikilol_structuraldensity.json` (< 1 MB)
* `wikilol_globaltaxonomy.json` (< 1 MB)
* `wikilol_sectionschema.json` (< 1 MB)
* `wikilol_sectionschema_exhaustive.csv` (6.96 MB)
* `WikiLoLParser_Log.json` (< 1 MB)

## Intended Utility
This dataset is designed for AI alignment research, network analysis, and structural engineering.
* **Spatial Reasoning Evaluation:** Providing a deterministic ground-truth map to test if language models can navigate conceptual hierarchies without relying on token-proximity memorization.
* **Domain-Specific Graph Theorization:** Utilizing the Subgraphs to train models on isolated domains without cross-domain contamination.

## Explicit Limitations & Structural Anomalies
* **Zero Narrative Text:** This dataset cannot be used to train Generative QA models. It contains only structural addresses and entity names.
* **Template Exclusions:** The underlying `mwparserfromhell` implementation strictly extracts standard Wikitables and list markup. Highly customized Wikipedia templates (e.g., massive legislative navboxes) are bypassed. This results in isolated "sink nodes" that possess massive incoming PageRank but zero outgoing edges. Researchers should account for this when normalizing adjacency matrices.

## Authorship
Extracted and compiled by Exorobourii LLC for independent research in core architectural efficiency and language model alignment.