ConceptNet

aid: conceptnet
name: ConceptNet
description: >-
  ConceptNet is a freely available multilingual knowledge graph that gives
  computers access to common-sense knowledge. It represents over 13 million
  links between concepts across 100+ languages, drawing from crowd-sourced
  resources (Open Mind Common Sense, Wiktionary), expert-created resources
  (WordNet, JMDict), and games with a purpose (Verbosity, nadya.jp). The
  public REST API provides JSON-LD responses and receives over 50,000 daily
  hits. ConceptNet also powers Numberbatch, a set of multilingual word
  embeddings aligned across languages that outperform word2vec, GloVe, and
  fastText on standard benchmarks.
url: https://conceptnet.io
image: https://conceptnet.io/img/conceptnet-logo.png
specificationVersion: '0.19'
created: '2026-06-13'
modified: '2026-06-13'
x-source: manual
x-category: Knowledge Graphs
tags:
  - Knowledge Graph
  - NLP
  - Semantic Web
  - Common Sense
  - Multilingual
  - Word Embeddings
  - Linked Data
  - Open Data
apis:
  - name: ConceptNet REST API
    description: >-
      The ConceptNet REST API exposes the full ConceptNet 5 knowledge graph
      via JSON-LD endpoints. Consumers can look up concept nodes by language
      and term, query edges by relation type, retrieve semantically related
      terms ranked by Numberbatch embedding similarity, compute pairwise
      relatedness scores between concepts, and normalize natural-language
      text into canonical ConceptNet URIs. No authentication or API key is
      required. Rate limits of 3,600 requests/hour and 120 requests/minute
      apply. The /related and /relatedness endpoints each count as 2 requests
      against the quota.
    humanURL: https://github.com/commonsense/conceptnet5/wiki/API
    baseURL: https://api.conceptnet.io
    tags:
      - Knowledge Graph
      - NLP
      - Semantic Relations
      - Multilingual
      - Common Sense
    properties:
      - type: Documentation
        url: https://github.com/commonsense/conceptnet5/wiki/API
      - type: GettingStarted
        url: https://github.com/commonsense/conceptnet5/wiki
      - type: Downloads
        url: https://github.com/commonsense/conceptnet5/wiki/Downloads
common:
  - type: Website
    url: https://conceptnet.io
  - type: Documentation
    url: https://github.com/commonsense/conceptnet5/wiki/API
  - type: GettingStarted
    url: https://github.com/commonsense/conceptnet5/wiki
  - type: GitHubOrganization
    url: https://github.com/commonsense
  - type: GitHubRepository
    url: https://github.com/commonsense/conceptnet5
  - type: License
    url: https://creativecommons.org/licenses/by-sa/4.0/
  - type: Downloads
    url: https://github.com/commonsense/conceptnet5/wiki/Downloads
  - type: FAQ
    url: https://github.com/commonsense/conceptnet5/wiki/FAQ
  - type: Support
    url: https://groups.google.com/g/conceptnet-users
  - type: Plans
    url: plans/conceptnet-plans-pricing.yml
  - type: RateLimits
    url: rate-limits/conceptnet-rate-limits.yml
  - type: FinOps
    url: finops/conceptnet-finops.yml
features:
  - name: Multilingual Knowledge Graph
    description: >-
      Covers concepts in 100+ languages with cross-language semantic links.
      Each concept node is identified by a URI such as /c/en/dog or /c/fr/chien,
      enabling cross-lingual knowledge transfer and multilingual NLP pipelines.
  - name: Semantic Relations
    description: >-
      Edges encode typed relations including IsA, UsedFor, CapableOf, AtLocation,
      Causes, HasProperty, PartOf, SimilarTo, Antonym, RelatedTo, and many more.
      Each edge carries a weight representing confidence from the source data.
  - name: JSON-LD Linked Data API
    description: >-
      All API responses use the JSON-LD format with @context, @id, and @type
      annotations, enabling seamless integration with RDF toolchains and
      semantic web applications.
  - name: Related Terms (Numberbatch Embeddings)
    description: >-
      The /related endpoint returns ranked semantically related concepts
      using ConceptNet Numberbatch word vectors — cross-lingual embeddings
      designed to avoid harmful stereotypes and outperform word2vec, GloVe,
      and fastText on analogy and similarity benchmarks.
  - name: Pairwise Relatedness Score
    description: >-
      The /relatedness endpoint returns a similarity score between 0 and 1
      for any two concept URIs, enabling quick semantic similarity checks
      without building a local embedding model.
  - name: Complex Edge Queries
    description: >-
      The /query endpoint accepts combinations of start, end, rel, node,
      other, and sources parameters to slice the knowledge graph by
      subject, object, relation type, or data source simultaneously.
  - name: URI Normalization
    description: >-
      The /uri endpoint converts raw natural-language text in any supported
      language into a canonical ConceptNet URI, handling tokenization,
      lowercasing, and language-specific normalization automatically.
  - name: No Authentication Required
    description: >-
      The public API requires no registration, API key, or OAuth token.
      Any HTTP client can query api.conceptnet.io directly.
useCases:
  - name: Semantic Similarity in NLP Pipelines
    description: >-
      Use /relatedness or /related to score or rank term similarity in
      question answering, text classification, and entity disambiguation
      tasks without training a custom embedding model.
  - name: Knowledge-Graph-Augmented AI
    description: >-
      Enrich LLM prompts or retrieval pipelines with structured commonsense
      facts by querying /c/{language}/{term} for edges describing causes,
      properties, and typical locations of a concept.
  - name: Cross-Language Information Retrieval
    description: >-
      Leverage ConceptNet's multilingual graph to expand a query in one
      language to synonymous concepts in another, supporting multilingual
      search and cross-lingual document clustering.
  - name: Educational Vocabulary Tools
    description: >-
      Build vocabulary-learning apps that surface semantic neighbors,
      antonyms, and example sentences for any word in dozens of languages
      using the IsA, SimilarTo, and Antonym relation edges.
  - name: Commonsense Reasoning Datasets
    description: >-
      Use ConceptNet as a gold-standard knowledge source for generating or
      evaluating commonsense reasoning benchmarks and training data for
      language models.
  - name: Chatbot Knowledge Enrichment
    description: >-
      Query ConceptNet for UsedFor, CapableOf, and AtLocation edges to
      give chatbots and virtual assistants grounded commonsense knowledge
      about everyday objects and actions.
integrations:
  - name: Python (conceptnet5)
    description: >-
      The open-source Python codebase includes a local AssertionFinder API
      for querying a self-hosted ConceptNet database without HTTP overhead.
    url: https://github.com/commonsense/conceptnet5
  - name: ConceptNet Numberbatch
    description: >-
      Pre-trained multilingual word embedding vectors (h5 and text formats)
      that can be loaded directly into NumPy, PyTorch, or TensorFlow for
      offline semantic similarity computation.
    url: https://github.com/commonsense/conceptnet-numberbatch
  - name: Linked Data / RDF
    description: >-
      JSON-LD responses integrate directly with JSON-LD processors, RDF
      triple stores (Apache Jena, Virtuoso), and SPARQL query engines.
solutions:
  - name: Zero-Cost Semantic Enrichment
    description: >-
      Query /related for any concept in an NLP pipeline at zero cost and
      without authentication, adding knowledge-graph context to otherwise
      purely statistical models.
  - name: Self-Hosted High-Volume Deployment
    description: >-
      For applications exceeding 3,600 requests/hour, download the full
      ConceptNet data dump and run a local PostgreSQL-backed instance using
      the open-source codebase, eliminating rate limits entirely.
maintainers:
  - FN: Kin Lane
    email: [email protected]