File size: 5,399 Bytes
14b5b33
4d3fd17
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
14b5b33
 
4d3fd17
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
14b5b33
4d3fd17
c7f7731
4d3fd17
c7f7731
4d3fd17
c7f7731
4d3fd17
c7f7731
4d3fd17
 
 
 
 
 
 
 
 
 
1
2
3
4
5
6
7
8
9
10
11
12
13
14
15
16
17
18
19
20
21
22
23
24
25
26
27
28
29
30
31
32
33
34
35
36
37
38
39
40
41
42
43
44
45
46
47
48
49
50
51
52
53
54
55
56
57
58
59
60
61
62
63
64
65
66
67
68
69
70
71
72
73
74
75
76
77
78
79
80
81
82
83
84
85
86
87
88
89
90
91
92
93
94
95
96
97
98
99
100
101
102
103
104
105
106
107
108
109
110
111
112
113
114
115
116
117
118
119
120
121
122
123
124
125
126
127
128
129
130
---
license: cc-by-4.0
pretty_name: ConceptNet 5 (Normalized SQLite, Filtered)
language:
  - en
  - fr
  - it
  - de
  - es
  - ar
  - fa
  - grc
  - he
  - la
  - hbo
multilinguality: multilingual
tags:
  - conceptnet
  - knowledge-graph
  - sqlite
  - normalized
---

# Normalized ConceptNet 5 (SQLite, Filtered)

This dataset contains a normalized, filtered, and optimized version of the ConceptNet 5.5 knowledge graph, ready for high-performance querying in a single SQLite file.

It is derived from the `cstr/conceptnet-de-indexed` dataset, which was a 23.6 GB un-normalized SQLite file containing 28.3 million nodes and 34 million edges.

This version has been processed to be significantly smaller, faster, and data-correct.

## Key Features

- **Normalized Schema**: The original 23.6 GB database stored massive text URLs (e.g., `http://conceptnet.io/c/en/dog`) in the 34M-row edge table. This version stores all 28M nodes in a `node_norm` lookup table and uses small, fast integer foreign keys (`start_fk`, `end_fk`) in the `edge_norm` table. This reduces the final database size by over 90%.

- **Targeted Language Filtering**: The 34 million edges have been filtered to keep only those relevant to a specific set of 11 languages: `en`, `fr`, `it`, `de`, `es`, `ar`, `fa`, `grc`, `he`, `la`, `hbo`.

- **Preserves Cross-Language Edges**: The filtering logic is data-safe. It keeps any edge where at least one of its nodes belongs to a target language. This is critical for preserving cross-lingual connections (e.g., a Japanese node `ja` linked to a German node `de`).

- **No Orphans**: The final `edge_norm` table links to the `node_norm` table. While `node_norm` contains all 28M original nodes (for lookup integrity), the `edge_norm` table only contains the filtered, relevant edges.

## Database Schema

This SQLite file (`conceptnet_normalized.db`) contains three tables:

### node_norm

- `node_pk` (INTEGER PRIMARY KEY): The new, unique integer ID for the node.
- `node_url` (TEXT UNIQUE NOT NULL): The original ConceptNet URL (e.g., `http://conceptnet.io/c/en/dog`).
- `language` (TEXT NOT NULL): The language code for the node (e.g., `en`, `de`), extracted from the source DB.

### rel_norm

- `rel_pk` (INTEGER PRIMARY KEY): The new, unique integer ID for the relation.
- `rel_url` (TEXT UNIQUE NOT NULL): The original relation URL (e.g., `http://conceptnet.io/r/IsA`).

### edge_norm

- `start_fk` (INTEGER NOT NULL): Foreign key to `node_norm.node_pk`.
- `end_fk` (INTEGER NOT NULL): Foreign key to `node_norm.node_pk`.
- `rel_fk` (INTEGER NOT NULL): Foreign key to `rel_norm.rel_pk`.
- `weight` (REAL NOT NULL): The edge weight.

## How to Use

You can query this database using any standard SQLite library.

```python
import sqlite3
import pandas as pd

DB_PATH = "conceptnet_normalized.db"  # Or path from hf_hub_download
conn = sqlite3.connect(f"file:{DB_PATH}?mode=ro", uri=True)

# Example: Get the top 5 'IsA' relationships for 'dog'
query = """
SELECT
    n_start.node_url AS start_node,
    r.rel_url AS relation,
    n_end.node_url AS end_node,
    e.weight
FROM edge_norm e
JOIN node_norm n_start ON e.start_fk = n_start.node_pk
JOIN node_norm n_end ON e.end_fk = n_end.node_pk
JOIN rel_norm r ON e.rel_fk = r.rel_pk
WHERE
    n_start.node_url = 'http://conceptnet.io/c/en/dog'
    AND r.rel_url = 'http://conceptnet.io/r/IsA'
ORDER BY e.weight DESC
LIMIT 5;
"""

df = pd.read_sql_query(query, conn)
print(df)

conn.close()
```

## Original Dataset Description

- **Homepage**: https://github.com/commonsense/conceptnet5/wiki
- **Repository**: https://github.com/commonsense/conceptnet5/wiki
- **Paper**: https://arxiv.org/abs/1612.03975

ConceptNet is a multilingual knowledge base, representing words and phrases that people use and the common-sense relationships between them. The knowledge in ConceptNet is collected from a variety of resources, including crowd-sourced resources (such as Wiktionary and Open Mind Common Sense), games with a purpose (such as Verbosity and nadya.jp), and expert-created resources (such as WordNet and JMDict).

This dataset is derived from the `conceptnet5` dataset (also on the Hub) and the `cstr/conceptnet-de-indexed` repository.

## Licensing Information

This 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.

The included data was created by contributors to Commonsense Computing projects, contributors to Wikimedia projects, DBPedia, OpenCyc, Games with a Purpose, Princeton University's WordNet, Francis Bond's Open Multilingual WordNet, and Jim Breen's JMDict.

For 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.

## Citation Information

If you use this data in your work, please cite the original ConceptNet 5.5 paper:

```bibtex
@inproceedings{speer2017conceptnet,
    author = {Robyn Speer and Joshua Chin and Catherine Havasi},
    title = {ConceptNet 5.5: An Open Multilingual Graph of General Knowledge},
    booktitle = {Proceedings of the AAAI Conference on Artificial Intelligence},
    year = {2017},
    pages = {4444--4451},
    url = {http://aaai.org/ocs/index.php/AAAI/AAAI17/paper/view/14972}
}
```