The Linked Data ecosystem and RDF as its graph data model have been around for many years already. Even though there is an increasing interest in knowledge graphs, many developers are scared off by RDF due to its reputation of being complicated. In this post, I will show some concrete examples on how RDF can be used in JavaScript applications, to illustrate that RDF is actually pretty easy to work with if you use the right tools.

To paraphrase Dan Brickley and Libby Miller, people think using RDF is hard because of its complexity. However, RDF is actually pretty simple, it merely allows you to handle very complex problems. To handle these problems, flexible and expressive tools are needed. These tools should be easy to use, so that we don’t get lost in the complexity of these problems during development. While there have been calls to make RDF easier, this post aims to show that RDF is not the problem, and that good tooling gets us where we want.

Concretely, this post covers the fundamentals of RDF in JavaScript, how to create RDF graphs yourself, how to retrieve them from the Web, and how to execute queries over them. I offer some concrete examples for each aspect using existing tools. It is not my aim to discuss all the available tooling, but instead to present the general idea, and to give pointers to tools that could be used for this.

Why JavaScript?

The RDF specification describes itself as “a framework for representing information in the Web”, which makes the Web an essential component in this story. The primary programming language for the Web is JavaScript, where it can be used for both client-side and server-side applications. Furthermore, JavaScript has become increasingly popular in the recent years, with the highly-performant V8 engine that can run JavaScript on clients and servers, the npm package manager, front-end frameworks such as React and Angular, TypeScript as a typed superset of JavaScript, and the continuously improving JavaScript language powered by the help of Babel. For these reasons and more, JavaScript is an ideal platform for RDF application development that offers a lot of flexibility.



Since 2018, the community has converged to a couple of specifications for enabling interoperability between different JavaScript applications, under the name of RDFJS. Today, most of the popular JavaScript libraries adhere to these specifications, which makes it possible to use them interchangeably, and in any combination. This allows you to for example use an RDF parser from one developer, and pipe its output into an RDF store from another developer.

The foundational part of RDFJS is its low-level data model specification, in which JavaScript interfaces are described for representing RDF terms and RDF quads. Five types of terms exist:

  • Named Node: Represents a thing by IRI, such as
  • Blank Node: Represents a thing without an explicit name.
  • Literal: Represents a raw value of a certain datatype, such as "Ruben" or 1992.
  • Variable: Represents a variable, which can be used for matching values within queries.
  • Default Graph: Represents the default graph in RDF. Other graphs can be represented with named or blank nodes.

RDF quads are defined as an object with RDF terms for subject, predicate, object and graph. An RDF triple is an alias of a quad, where the graph is set to the default graph. For the remainder of this document, I will just refer to RDF quads.

Finally, a Data Factory interface is defined, which allows you to easily create terms and quads that conform to this interface. Different Data Factory implementations exist, such as rdf-data-factory and the factory from N3.js. For example, creating a quad for representing someone’s name with a data factory can be done like this:

import { DataFactory } from 'rdf-data-factory';

const factory = new DataFactory();

const quad = factory.quad(
  factory.namedNode(''), // subject
  factory.namedNode(''),          // predicate
  factory.literal('Ruben')                              // object

Reading raw values from the quad can be done as follows:

quad.subject.value === '';
quad.predicate.value === '';
quad.object.value === 'Ruben';

For checking whether or not quads and terms are equal to each other, the equals method can be used:

factory.literal('Ruben').equals(factory.literal('Ruben'));  // true
factory.literal('Ruben').equals(factory.literal('Ruben2')); // false
quad.equals(quad); // true

Next to the RDFJS data model, separate specifications exist for handling RDF streams and high-level datasets. An overview of all JavaScript libraries that support any of the RDFJS specifications can be found on For most of these specifications, corresponding TypeScript typings exist, and many libraries ship with their own typings as well, which makes RDFJS especially useful if you want to develop more strongly-typed JavaScript applications.

While there are general-purpose RDF(JS) libraries available such as rdflib and rdf, I will focus on dedicated tools with scoped functionality for the remainder of this post. Since they all conform to the RDFJS interfaces, they can however be used interchangably.

Creating RDF graphs

Now that we can create RDF quads, we can start combining them into RDF graphs to represent information and knowledge. The easiest way to create an RDF graph is using an in-memory datastructure. N3.Store is an efficient in-memory index for storing RDF, and doing quad pattern-based lookups.

Adding Quads

RDFJS quads can be added to the store by calling the addQuad method:

const N3 = require('n3');
const store = new N3.Store();


If you have multiple quads to insert, then the array-based addQuads may be used instead:

// Array-based

If your quads originate are produced as a stream (like from a parser), then the stream-based import method may be used instead:

const rdfParser = require("rdf-parse").default;
const quadStream = rdfParser.parse(fs.createReadStream('cartoons.ttl'),
  { contentType: 'text/turtle' });

  .on('end', () => console.log('Stream has been imported'));

In cases you quickly need to create a new store for a quad stream, then rdf-store-stream.js can be used as a convenience tool:

const storeStream = require("rdf-store-stream").storeStream;

const store = await storeStream(quadStream);

Looking Up Quads

Looking up quads can be done through quad pattern-matching using the match method from the RDFJS Source interface, where any of the quad components can be left undefined (or null) to allow anything to match with it. Since real-world RDF graphs are typically very large, lookups typically happen asynchronously so that matched quads can be processed in a memory-efficient manner as soon as they are discovered. For this, the match method returns a stream of RDF quads. This stream is compatible with the Readable stream interface of Node.js.

For example, looking up everything with as subject can be done like this:

const quadStream = store.match(
  // predicate, object, graph arguments are left undefined
  .on('error', console.error)
  .on('data', (quad) => {
    // Handle our quad...
  .on('end', () => console.log('Done!'));

Looking up all the names of can similarly be done like this:

const quadStream2 = store.match(

Resulting quads streams can be used for various things, such as creating a new N3.Store, or serializing it to a Turtle file:

  .pipe(new N3.StreamWriter())

If your application does not require stream-based processing, and you just want to retrieve an array of quads, then the getQuads method may be used instead:

const quadsArray = store.getQuads(
for (const quad of quadsArray) {
    // Handle our quad...

Alternatively, quad streams can be converted to arrays using tools such as arrayify-stream.

Even though N3.Store uses a highly-efficient four-level index to store and lookup RDF quads, its main restriction is that it stores everything in-memory. On average, storing ten million triples requires around 1GB of memory, which should still be achievable on average hardware. However, one hundred million triples may already be too much for average harware (~10GB). For these cases, dedicated on-disk tools may be preferable, such as the LevelDB-based node-quadstore, or via highly compressed read-only HDT files (Node.JS bindings).

Next to these low-level quad stores, more high-level datastructures exist, such as the Dataset interface. This interface offers more convenience features such as array-like operations, calculating set algebra operations and canonicalization, at the cost of loosing asynchronicity. Implementations of this Dataset interface exist in graphy.js and rdf-dataset-indexed.

Dereferencing RDF graphs

Thanks to the Linked Data rules, RDF is available on the Web behind HTTP URIs, such as or These URIs can be dereferenced to look up its contents. Since RDF on the Web exists in various RDF serializations such as Turtle, RDFa, and JSON-LD, the dereferencing of RDF involves picking the correct parser for the returned RDF serialization.

Concretely client-side RDF dereferencing for a given URI typically involves the following steps:

  1. Determine the media types from the available RDF parsers, such as text/n3 or text/html.
  2. Order the media types by priority, as different parsers may have different performance requirements.
  3. Perform content-negotation for the given URI to retrieve an appropriate RDF serialization from the server.
  4. Pick the appropriate RDF parser based on the server’s returned content type.

If you want to retrieve RDF from the Web, then these same steps have to be followed every time again. As such, different tools are available to hide this process so that RDF dereferencing becomes significantly easier. rdf-dereference.js is a tool that does exactly that. It accepts a URI, and it outputs an stream of RDF quads:

const rdfDereferencer = require("rdf-dereference").default;

const { quads: quadSream } = await rdfDereferencer
quadSream.on('data', (quad) => console.log(quad))
     .on('error', (error) => console.error(error))
     .on('end', () => console.log('Done!'));

rdf-dereference.js has support for all major RDF serializations using fully spec-compliant streaming parsers, and prioritizes them by parsing efficiency. Alternatively, if you need more configuration flexibility regarding for instance the serializations you want to support, then @rdfjs/fetch-lite can be used. If you want to take care of HTTP fetching yourself, and use something more low-level for just parsing in any serialization, then the rdf-parse.js can be used.

Querying RDF graphs

While the previous two sections discussed the low-level handling of RDF on quad-level, this section focuses more on the linked aspect of RDF data. Concretely, I discuss some techniques for querying through RDF graphs by looking up information across multiple quads.

The process of querying is typically handled by a query engine that accepts a query in some kind of declarative language, and returns the results in a certain format. While there are several techniques out there to query RDF, I will focus on three query languages that have different goals:

  • LDflex: A JavaScript-based domain-specific language for simple writing path-based expressions.
  • GraphQL-LD: An extension of the GraphQL query language for querying Linked Data.
  • SPARQL: The standard for querying RDF as recommended by the World Wide Web Consortium.

These three languages differ in terms of their query expressivity (the variety of possible queries) and developer complexity (how easy queries can be written). While LDflex is the easiest to use, it offers the least expressive power. SPARQL on the other hand offers the most expressivity, but is typically harder to use. GraphQL-LD can be seen a trade-off between both. Depending on the application, any of these can be used, even in combination. Hereafter, I give some examples for each of them.

Query expressivity and developer complexity


When you need to look up data that can be represented by a single chain of properties, then LDflex is probably the easiest way to do this. LDflex has been designed to look like the traversal of a JavaScript object, while instead this traversal is internally translated to a SPARQL query using a JSON-LD context. This SPARQL query is then delegated to a third-party query engine such as Comunica for execution.

For example, finding the names of all the friends of that are listed in can be done as follows:

const { PathFactory } = require('ldflex');
const { default: ComunicaEngine } = require('ldflex-comunica');

// Setup path expression
const context = {
  "@context": {
    "@vocab": "",
    "friends": "knows",
const paths = new PathFactory({
	queryEngine: new ComunicaEngine(''),
const person = paths.create({ subject: '' });

// Execute expressions
(async function() {
  console.log(await + ' is friends with:');
  for await (const name of person.friends.givenName)
    console.log('- ' + name);

Similar tools like this exists for traversing local RDF graphs instead of remote graphs, such as rdf-object.js, simplerdf and clownface.


If single chained properties are not sufficient and you need to retrieve multiple values of things, then a tree-based query language as provided by GraphQL may be more appropriate. It allows multiple values to be retrieved via a single query, instead of having to write several separate ones.

GraphQL-LD is a technique that allows Linked Data to be queried using GraphQL queries, by enhancing it with a JSON-LD context, similar to LDflex. Internally, queries are also translated to SPARQL and delegated to a third-party query engine.

Below, you can see an example on how you can retrieve my name, image and friends using a single query:

import {Client} from "graphql-ld";
import {QueryEngineComunica} from "graphql-ld-comunica";

// Setup the client
const context = {
  "@context": {
    "Ruben": "",
    "foaf": "",
    "name": "foaf:name" ,
    "image": "foaf:img",
    "friends": "foaf:knows" 
const client = new Client({
  queryEngine: new QueryEngineComunica({
    sources: [''],

// Define and execute a query
const query = `{
  name @single
  image @single
  friends {
    name @single
const { data } = await client.query({ query });

The returned data will resemble the query structure as follows:

  "name": "Ruben Taelman",
  "image": "",
  "friends": [
    { "name": "Anastasia Dimou" },
    { "name": "Ben De Meester" },
    { "name": "Brecht Van de Vyvere" },


When neither single path expressions or tree-based queries are sufficient for you use case, then the highly-expressive SPARQL query language probably will meet your needs. At the cost of a more complex language, it offers many features such as graph-based pattern matching, filtering, aggregation, negation, optionals, subqueries, constructing RDF triples, and more.

Comunica is a query engine platform for Web querying in JavaScript. It is not a query engine by itself, but a framework to build query engines. Due to its flexibility and configurability, it can be used for a variety of use cases. This is also why it is used internally for the underlying query engines in LDflex and GraphQL-LD.

@comunica/actor-init-sparql is one example of a Comunica engine that has been built to support SPARQL queries. It can be used as follows to lookup my name and ten of my friends:

const newEngine = require('@comunica/actor-init-sparql').newEngine;
const myEngine = newEngine();

const config = { sources: [''] };
const query = `
PREFIX foaf: <>
  <> foaf:name ?name.
    SELECT ?friendName WHERE {
      <> foaf:knows [
          foaf:name ?friendName
    } LIMIT 10

const { bindingsStream } = await myEngine.query(query, config);
bindingsStream.on('data', (data) => console.log(data.toObject()));
bindingsStream.on('end', () => console.log('Done!'));

Note that the configuration object supports multiple sources, which enables queries across multiple datasets. This allows you to for example find the common interests of Ruben Verborgh and myself, and fetch the interest labels from DBpedia:

const config = { sources: [
] };
const query = `
PREFIX rdfs: <>
PREFIX foaf: <>
SELECT ?interestName
  ruben:me foaf:topic_interest ?interest.
  rubent:me foaf:topic_interest ?interest.
  ?interest rdfs:label ?interestName.
  FILTER LANGMATCHES(LANG(?interestName),  "EN")

Next to querying over URI-based sources, it is also possible to query over local sources or in-memory RDFJS sources.

Future RDF Tooling

While we have come a long way for handling RDF in JavaScript, technology always keeps evolving. In the future, we will most likely see a shift towards more query-driven applications, where query engines gain more responsibility for handling complex tasks. This will allow applications to define their queries and sources, and query engines will take care of all the complexities for resolving the query, such as query optimization, source selection, resulting joining, authentication against sources, traversal of links, reasoning, and more. Current query engines can already perform some of these tasks, but more research and development effort is needed for things such as source selection and link traversal, which are part of my personal research and development goals for the upcoming years.