cstr/conceptnet_normalized

1---2title: Normalized ConceptNet Explorer3emoji: ⚡4colorFrom: green5colorTo: blue6sdk: gradio7sdk_version: 5.49.18app_file: app.py9pinned: true10license: cc-by-sa-4.011tags:12  - conceptnet13  - knowledge-graph14  - sqlite15  - normalized16  - gradio17  - fast-queries18---19 20# ⚡ Normalized ConceptNet Explorer (V7)21 22This application is a high-performance explorer for a normalized, filtered, and optimized version of the ConceptNet 5.5 knowledge graph.23 24It is designed to be **extremely fast**, returning queries in milliseconds instead of minutes. It queries a 1.78 GB optimized SQLite database with integer-based joins, not the 23.6 GB un-normalized file.25 26## Features27 28This app provides a full suite of tools to explore the normalized database:29 30- **⚡ Semantic Profile**: Explore relations for any word in real-time. This now runs in ~4 fast SQL queries instead of 24+ slow ones.31- **⚡ Query Builder**: Build custom queries (start node, relation, end node) that are executed with fast, integer-based joins.32- **⚡ Raw SQL**: Execute SQL queries directly against the new, normalized database schema (see schema below).33- **⚡ Schema**: Browse the new, efficient database schema, including all tables, indexes, and row counts.34 35## How It Works: The Normalized Database36 37This app's speed and correctness come from the new database it queries: [cstr/conceptnet-normalized-multi](https://huggingface.co/datasets/cstr/conceptnet-normalized-multi).38 39This database was created by a V7 normalization script that fixed critical issues found in the original data:40 411. **Normalization (Speed & Size)**: The original 23.6 GB `edge` table (34M rows) was bloated with text URLs. The new 1.78 GB `edge_norm` table replaces these with tiny integer foreign keys.42 432. **Data Correctness (V7 Fix)**: The original `node` table (28M rows) was used as the source of truth. We migrated all 28M nodes and their authoritative `language` columns.44 453. **Preserves Cross-Language Links**: The 34M edges were filtered to keep any edge where at least one node (start or end) was in our 11 target languages (`en`, `de`, `fr`, `it`, `es`, `ar`, `fa`, `grc`, `he`, `la`, `hbo`). This is critical, as it correctly preserves cross-language connections (e.g., `犬 (ja) -> hund (de)`), which were broken in previous attempts.46 47The result is a clean, fast, and data-correct database that contains all relevant connections for our target languages.48 49## Supported Languages50 51This normalized version includes edges for 11 languages:52- English (en)53- German (de)54- French (fr)55- Italian (it)56- Spanish (es)57- Arabic (ar)58- Persian (fa)59- Ancient Greek (grc)60- Hebrew (he)61- Latin (la)62- Biblical Hebrew (hbo)63 64Cross-language connections from other languages to these target languages are preserved.65 66## Original Dataset Information67 68This work includes data from ConceptNet 5, which was compiled by the Commonsense Computing Initiative. ConceptNet 5 is freely available under the Creative Commons Attribution-ShareAlike license (CC BY SA 4.0) from http://conceptnet.io.69 70For a full list of licenses and attributions for included resources such as WordNet, Open Multilingual WordNet, and Wikimedia projects, please see the original dataset card.71 72## Citation Information73 74If you use this data in your work, please cite the original ConceptNet 5.5 paper:75 76```bibtex77@inproceedings{speer2017conceptnet,78    author = {Robyn Speer and Joshua Chin and Catherine Havasi},79    title = {ConceptNet 5.5: An Open Multilingual Graph of General Knowledge},80    booktitle = {Proceedings of the AAAI Conference on Artificial Intelligence},81    year = {2017},82    pages = {4444--4451},83    url = {http://aaai.org/ocs/index.php/AAAI/AAAI17/paper/view/14972}84}85```