ConceptNet is a semantic network containing lots of things computers should know about the world, especially when understanding text written by people.
It is built from nodes representing concepts, in the form of words or short phrases of natural language, and labeled relationships between them. These are the kinds of things computers need to know to search for information better, answer questions, and understand people's goals. If you wanted to build your own Watson, this should be a good place to start!
Notice how the relations between concepts can be abstract notions such as MadeOf, which we use to mean the same thing across all languages; or they can be language-specific text such as "can cross".
You can click any of these links, or use the search bar above, to begin browsing ConceptNet.
API and Documentation
The newest release of ConceptNet, ConceptNet 5.2, is documented on our wiki.
The documentation includes how to use our REST API, which allows you to:
- Retrieve the data for particular nodes and edges
- Query for edges with given properties
- Measure and query the semantic distance between nodes
ConceptNet 5 is free
ConceptNet 5 comes largely from the hard work of hundreds of thousands of people who gave their time and knowledge for free. So ConceptNet is free as well, released under a choice of two Creative Commons licenses:
- You can get the entirety of ConceptNet 5 under the Creative Commons Attribution-ShareAlike 3.0 license.
- You may also use a smaller version, called the "ConceptNet 5 Core", under the Creative Commons Attribution 3.0 license. This version is free for any purpose as long as you give credit to the Digital Intuition team. However, this version is necessarily missing a large number of statements learned from Wikipedia, Wiktionary, and DBPedia, all of which are Attribution-ShareAlike resources.
(Until we have a separate download, you get ConceptNet 5 Core by using
ConceptNet 5 and discarding all nodes with
Sources and how to contribute
Previous versions of ConceptNet were a home-grown crowd-sourced project, where we ran a Web site collecting facts from people who came to the site. The Web of Data is much bigger than that now. Our data comes from many different sources, many of which you can contribute to and improve not just the state of computational knowledge, but of human knowledge.
- To begin with, ConceptNet 5 contains almost all the data from ConceptNet 4, created by contributors to the Open Mind Common Sense project.
- Much of our knowledge comes from the English Wikipedia and its contributors, through two sources:
- DBPedia extracts knowledge from the infoboxes that appear on articles.
- ReVerb is a machine reading project, extracting relational knowledge from the actual text of each article.
- We have also parsed a large amount of content from the English Wiktionary, including synonyms, antonyms, translations of concepts into hundreds of languages, and multiple labeled word senses for many English words.
- More dictionary-style knowledge comes from WordNet.
- Some knowledge about people's intuitive word associations comes from "games with a purpose". We learn things in English from the GWAP project's word game Verbosity, and in Japanese from nadya.jp.
ConceptNet 5 is a graph
To be precise, it's a hypergraph, meaning it has edges about edges. Each statement in ConceptNet has justfications pointing to it, explaining where it comes from and how reliable the information seems to be.
Previous versions of ConceptNet has been distributed as idiosyncratic database structures plus some software to interact with them. ConceptNet 5 is not a piece of software or a database; it is a graph. It's a set of nodes and edges, which we can represent in multiple formats including JSON. You probably know better than we do what software you want to use to interact with it!
(That said, you can have our idiosyncratic Solr index if you want, but that's not ConceptNet, it's just a system for quickly looking things up in ConceptNet.)
Some other interesting properties:
- The ConceptNet graph is ID-less. Every node and assertion contains all the information necessary to identify it and no more in its URI, and does not rely on arbitrarily-assigned IDs. The advantage of this is that if multiple branches of ConceptNet are developed in multiple places, we can later merge them simply by taking the union of the nodes and edges. (And we hope for this to happen!)
- ConceptNet supports linked data: you can download a list of links to the greater Semantic Web, via DBPedia and via RDF/OWL WordNet. For example, our concept cat is linked to the DBPedia node at http://dbpedia.org/resource/Cat.
Downloading ConceptNet 5
If you want all the data in ConceptNet for your application, you can have it! We provide the data in three formats:
- Flat JSON: files in which each line is a ConceptNet edge in JSON format
- Solr JSON: specially formatted JSON files that can be loaded into an Apache Solr index for quick retrieval
- CSV: simple tabular data, very convenient for command-line searching or as input to machine learning
ConceptNet 5 is an open-source project, developed with GPLv3 code hosted on GitHub.
ConceptNet is part of the Commonsense Computing Initiative, a collaboration between the MIT Media Lab and other labs and companies around the world. If you want to set up a sister project to collect a specific kind of data -- perhaps focusing on a particular domain or improving our coverage of a particular language -- we'd be happy for you to contact us.
The development of ConceptNet 5 is led by Rob Speer, advised by Catherine Havasi. It is also developed by Julian Chaidez, Justin Venezuela, and Yen-Ling Kuo.
Mailing list and contact information
For general questions and further information, join our mailing list on Google Groups.
For specific inquiries about our research group, e-mail firstname.lastname@example.org.