© Kemal Şanlı

An open-source named entity visualiser for the modern web

· by Ines Montani

Named Entity Recognition is a crucial technology for NLP. Whatever you're doing with text, you usually want to handle names, numbers, dates and other entities differently from regular words. To help you make use of NER, we've released displaCy-ent.js. This post explains how the library works, and how to use it.

displaCy ENT is a lightweight and extensible library that fetches JSON-formatted named entity annotations and transforms them into semantic HTML. This way, the original text stays intact and we can rely on web standards and avoid extra dependencies. The Brat visualiser for example is great and comes with a lot of options, but it also requires a 500+ line stylesheet, three fonts and a total of 11 JavaScript dependencies, including jQuery with two plugins. This can be a bit heavy if you only want to show a simple, static visualisation, or implement a custom workflow.

Here's an example of a text rendered by displaCy ENT:

Isn't PERSON the wrong label for 'Recode'?Yes — but that also makes it a good example. No model is perfect. This example is hard because the entity hasn't been seen in the training data, and "interview with PERSON" is more common than "interview with ORG".

The labels are displayed and styled using only CSS selectors — and you can do some pretty powerful stuff with that. You'll see a few examples later on.

Using displacy-ent.js in your projects

To use displaCy ENT, include displacy-ent.js and initialize a new instance specifying the API and settings. The parse(text, model, ents) method renders a parse generated by spaCy's services, which you can download and run for free.A note on compatibilitydisplaCy ENT is written in ECMAScript 6. For full, cross-browser compatibility, make sure to use a compiler like Babel. For more info, see this compatibility table.

// Your API
const api = 'http://localhost:8000';

// Init displaCY ENT
const displacy = new displaCyENT(api, {
    container: '#displacy'

const text = 'When Sebastian Thrun started working on self-driving cars at Google in 2007, few people outside of the company took him seriously.';
const model = 'en';

// Entities to visualise
const ents = ['person', 'org', 'date'];

// Parse text
displacy.parse(text, model, ents);

For a full list of available settings, see the Readme. Alternatively, you can also use render(text, spans, ents) to manually render a JSON-formatted set of entity spans returned from our REST service.

Example: Using the static visualisation on your site

If you're only looking to visualise the entities of a given text, there's no need to include any JavaScript at all. Simply head over to our demo and copy-paste the HTML and CSS. CodePen is a great tool to test this and see it in action immediately. displaCy ENT also logs the full HTML markup to the console.

Example: Dynamically generate a visualisation

Let's say you want to add entity visualisation to a bunch of text on your site. You can do this by replacing the content of your container with the visualised output of displaCy ENT once the document has loaded:

Usage exampleconst displacy = new displaCyENT('http://localhost:8000', {
    container: '#displacy',
    defaultModel: 'en',
    defaultEnts: ['person', 'org', 'date']

// Replace text in container with visualisation once document has loaded
document.addEventListener('DOMContentLoaded', (ev) => {
    const text = document.querySelector('#displacy').textContent;

Note that this only grabs the textContent and ignores line breaks, paragraphs and other HTML tags. As a workaround, you could replace line beaks and pargraphs with newlines in the container's innerHTML and then grab its textContent:

Hacky workaroundconst container = document.querySelector('#displacy');
container.innerHTML = container.innerHTML
                      .replace(/<br\s*[\/]?>/gi, '\n')
                      .replace(/<p[^>]*>/g, '')
                      .replace(/<\/p>/g, '\n\n');

Stripping HTML in JSIt might be tempting to add another regular expression to strip out all HTML tags in the first step. Don't do this. Regular expressions are not suitable to parse arbitrary HTML. But if you must, only do it on your content that you control.

In general, I'd recommend to only use displaCy ENT on plain, unformatted text. If you're working with user input or other unpredictable content, opt for a library to sanitise HTML and prevent XSS.

How displaCy ENT works

displaCy ENT wraps the entities in the HTML5 <mark> element for highlighted text. Each entity is assigned the data attribute data-entity:

Example markup<div class="entities">
    When <mark data-entity="person">Sebastian Thrun</mark> started working on self-driving
    cars at <mark data-entity="org">Google</mark> in <mark data-entity="date">2007</mark>,
    few people outside of the company took him seriously.

This is done using a simple JavaScript function. It takes the list of spans returned by the server, slices the original text into fragments and, if an entity should be visualised, wraps it in a <mark> tag and adds the right data attribute. The data attributes are then displayed after the entity in CSS using the :after pseudo-element:

Example CSS/* Display value of data-entity after entity */
[data-entity]:after {
    content: attr(data-entity);
    display: inline-block;
    vertical-align: middle;
    /* other visual styling */

/* Style "person" entity */
[data-entity="person"] {
    background: rgb(166, 226, 45);

[data-entity="person"]:after {
    background: rgba(166, 226, 45, 0.2);
    border-color: rgb(166, 226, 45);

With CSS, you have almost infinite possibilities to present the entities. Here are a few examples with different colours, borders and hover effects:

There's another cool thing you can do with advanced selectors: You can make it only show the entity label for the first instance, or rather, hide all other ones and only show them on hover:

The trick here is the element ~ element selector that selects all elements following another element. Basically, we're adding the label to all entities and then hiding it for all entities with data-entity="person" following another one, i.e. all except for the first.What about :first-of-type?My first instinct was to use the :first-of-type pseudo-class — but this doesn't work here, as it only selects for elements and not attributes. The ~ selector hack is currently the only way to make attribute queries like this possible.

CSS Markup (Excerpt)[data-entity="person"] {
    transition: opacity 0.25s ease;

[data-entity="person"] ~ [data-entity="person"]:after {
    opacity: 0;

[data-entity="person"] ~ [data-entity="person"]:hover:after {
    opacity: 1;

This needs to be done for all occuring entity types. In my example, I used a simple Sass loop to automate this:

Sass loop (Excerpt)$entities: ( person: #f44d61, org: #8e7dff, date: #2fbbab )

@each $entity, $color in $entities
    [data-entity="#{$entity}"] ~ [data-entity="#{$entity}"]
            opacity: 0

            opacity: 1

What's next?

In the future, we're planning to add support for more annotation formats like Google's NLP API and CoreNLP. I'd also love to release more CSS themes for different entity styles. If you'd like to contribute one, check out the current Sass template here and feel free to submit a pull request.

Ines Montani
About the author

Ines Montani

Ines is a co-founder of Explosion and a core developer of the spaCy NLP library and the Prodigy annotation tool. She has helped set a new standard for user experience in developer tools for AI engineers and researchers.