README.md 2.87 KB
Newer Older
Amirouche's avatar
cosmit    
Amirouche committed
1
2
3
4
5
# ConceptNet


## Overview

6
7
8
ConceptNet aims to give computers access to common-sense knowledge, the kind of
information that ordinary people know but usually leave unstated.

Amirouche's avatar
cosmit    
Amirouche committed
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
ConceptNet is a semantic network that represents things that computers
should know about the world, especially for the purpose of
understanding text written by people. Its "concepts" are represented
using words and phrases of many different natural language -- unlike
similar projects, it's not limited to a single language such as
English. It expresses over 13 million links between these concepts,
and makes the whole data set available under a Creative Commons
license.

Much of the current development of ConceptNet involves using it as an
input for machine learning about the semantics of text. Its
multilingual representation makes it particularly expressive, because
the semantic overlaps and differences between languages are a useful
signal that a learning system can learn from.

ConceptNet grew out of Open Mind Common Sense, an early project for
crowd-sourced knowledge, and expanded to cover many different
languages through a collaboration with groups around the
world. ConceptNet is cited in many research papers, and its public API
gets over 50,000 hits per day.


This Python package contains a toolset for building the ConceptNet 5
knowledge graph, possibly with your own custom data, and it serves the
HTML interface and JSON Web API for it.
34

Rob Speer's avatar
Rob Speer committed
35
You don't need this package to simply access ConceptNet 5; see
Amirouche's avatar
cosmit    
Amirouche committed
36
37
http://conceptnet.io for more information and a browsable Web
interface with an API.
38

39
Further documentation is available on the [ConceptNet wiki][].
Rob Speer's avatar
Rob Speer committed
40

Amirouche's avatar
cosmit    
Amirouche committed
41
42
Licensing and attribution appear in `LICENSE.txt` and
`DATA-CREDITS.md`.
Rob Speer's avatar
Rob Speer committed
43
44
45
46


## Discussion groups

Amirouche's avatar
cosmit    
Amirouche committed
47
48
49
If you're interested in using ConceptNet, please join the
conceptnet-users Google group, for questions and occasional
announcements: http://groups.google.com/group/conceptnet-users?hl=en
Rob Speer's avatar
Rob Speer committed
50

Amirouche's avatar
cosmit    
Amirouche committed
51
52
For real-time discussion, ConceptNet also has a chat channel on
Gitter: https://gitter.im/commonsense/conceptnet5
Rob Speer's avatar
Rob Speer committed
53
54
55
56


## Installing and building ConceptNet

Amirouche's avatar
cosmit    
Amirouche committed
57
58
59
60
To be able to run all steps of the ConceptNet build process, you'll
need a Unix command line (Ubuntu 16.04 works great), Python 3.5 or
later, 30 GB of RAM, and some other dependencies. See the [build
process][] on our wiki for instructions.
Rob Speer's avatar
Rob Speer committed
61

Amirouche's avatar
cosmit    
Amirouche committed
62
63
You may not need to build ConceptNet yourself! Try the [Web API][]
first.
Rob Speer's avatar
Rob Speer committed
64

65
66
67
[build process]: https://github.com/commonsense/conceptnet5/wiki/Build-process
[Web API]: https://github.com/commonsense/conceptnet5/wiki/API
[ConceptNet wiki]: https://github.com/commonsense/conceptnet5/wiki
Robyn Speer's avatar
Robyn Speer committed
68
69
70
71


## Testing

Amirouche's avatar
cosmit    
Amirouche committed
72
73
Run `pytest` to test the ConceptNet libraries and a small version of
the build process.
Robyn Speer's avatar
Robyn Speer committed
74

Amirouche's avatar
cosmit    
Amirouche committed
75
76
Run `pytest --quick` to re-run the tests more quickly, with the
assumption that the small test database has already been built.
Robyn Speer's avatar
Robyn Speer committed
77

Amirouche's avatar
cosmit    
Amirouche committed
78
79
Run `pytest --fulldb` to run additional tests on the fully built
ConceptNet database.