Word clouds, which are images showing scattered words in different sizes, are a popular way to visualise large amounts of text. Words that appear more frequently in the given text are larger, and less common words are smaller or not shown at all.
In this tutorial, we’ll build a web application using Python and Flask that transforms the latest news stories into word clouds and displays them to our visitors.
At the end of this tutorial, our users will see a page similar to the one shown below, but containing the latest news headlines from BBC news. We’ll learn some tricks about web scraping, RSS feeds, and building image files directly in memory along the way.
We’ll be building a simple web application step-by-step and explaining each line of code in detail. To follow, you should have some basic knowledge of programming and web concepts, such as what if statements are and how to use URLs. We’ll be using Python for this tutorial, but we won’t assume that you’re a Python expert.
Specifically, we’ll:
We’ll be using the online programming environment Repl.it so you won’t need to install any software locally to follow along step by step. If you want to adapt this guide to your own needs, you should create a free account by going to repl.it and follow their sign up process.
We previously looked at basic web scraping in an introduction to web scraping. If you’re completely new to the idea of automatically retrieving content from the internet, have a look at that tutorial first.
In this tutorial, instead of scraping the links to news articles directly from the BBC homepage, we’ll be using RSS feeds - an old but popular standardised format that publications use to let readers know when new content is available.
RSS feeds are published as XML documents. Every time BBC (and other places) publishes a new article to their home page, they also update an XML machine-readable document at http://feeds.bbci.co.uk/news/world/rss.xml. This is a fairly simple feed consisting of a <channel> element, which has some metadata and then a list of <item> elements, each of which represents a new article. The articles are arranged chronologically, with the newest ones at the top, so it’s easy to retrieve new content.
If you click on the link above, you won’t see the XML directly. Instead, it has some associated styling information so that most web browsers will display something that’s a bit more human friendly. For example, opening the page in Google Chrome shows the page below. In order to view the raw XML directly, you can right-click on the page and click “view source”.
RSS feeds are used internally by software such as the news reader Feedly and various email clients. We’ll be consuming these RSS feeds with a Python library to retrieve the latest articles from BBC.
In this tutorial, we’ll be building our web application using Repl.it, which will allow us to have a consistent code editor, environment, and deployment framework in a single click. Head over there and create an account. Choose to create a Python Repl, and you should see an editor where you can write and run Python code, similar to the image below. You can write Python code in the middle pane, run it by pressing the green “run” button, and see the output in the right pane. In the left pane, you can see a list of files, with main.py added there by default.
In the previous webscraping tutorial we used BeautifulSoup to look for hyperlinks in a page and extract them. Now that we are using RSS, we can simply parse the feed as described above to find these same URLs. We will be using the Python feedparser library to do this.
Let’s start by simply printing out the URLs for all of the latest articles from BBC. Switch back to the main.py file in the Repl.it IDE and add the following code.
import feedparser
BBC_FEED = "http://feeds.bbci.co.uk/news/world/rss.xml"
feed = feedparser.parse(BBC_FEED)
for article in feed['entries']:
print(article['link'])Feedparser does most of the heavy lifting for us, so we don’t have to get too close to the slightly cumbersome XML format. In the code above, we parse the feed into a nice Python representation (line 4), loop through all of the entries (the <item> entries from the XML we looked at earlier), and print out the link elements.
If you run this code, you should see a few dozen URLs output on the right pane, as in the image below.
We don’t just want to print this data out in the Repl console. Instead, our application should return information to anyone who uses a web browser to visit our application. We’ll, therefore, install the lightweight web framework Flask and use this to serve web content to our visitors.
In the main.py file, we need to modify our code to look as follows:
import feedparser
from flask import Flask
app = Flask(__name__)
BBC_FEED = "http://feeds.bbci.co.uk/news/world/rss.xml"
@app.route("/")
def home():
feed = feedparser.parse(BBC_FEED)
urls = []
for article in feed['entries']:
urls.append(article['link'])
return str(urls)
if __name__ == '__main__':
app.run('0.0.0.0')Here we still parse the feed and extract all of the latest article URLs, but instead of printing them out, we add them to a list (urls), and return them from a function. The interesting parts of this code are:
/).Press “run” again, and you should see a new window appear in the top right pane. Here we can see a basic web page (viewable already to anyone in the world by sharing the URL you see above it), and we see the same output that we previously printed to the console.
The URLs aren’t that useful to us, as we eventually want to display a summary of the content of each URL. The actual text of each article isn’t included in the RSS feed that we have (some RSS feeds contain the full text of each article), so we’ll need to do some more work to download each article. First, we’ll add the third-party libraries requests and BeautifulSoup as dependencies, again just using the “magic import”. We’ll be using these to download the content of each article from the URL and strip out extra CSS and JavaScript to leave us with plain text.
Now we’re ready to download the content from each article and serve that up to the user. Modify the code in main.py to look as follows.
import feedparser
import requests
from flask import Flask
from bs4 import BeautifulSoup
app = Flask(__name__)
BBC_FEED = "http://feeds.bbci.co.uk/news/world/rss.xml"
LIMIT = 2
def parse_article(article_url):
print("Downloading {}".format(article_url))
r = requests.get(article_url)
soup = BeautifulSoup(r.text, "html.parser")
ps = soup.find_all('p')
text = "\n".join(p.get_text() for p in ps)
return text
@app.route("/")
def home():
feed = feedparser.parse(BBC_FEED)
article_texts = []
for article in feed['entries'][:LIMIT]:
text = parse_article(article['link'])
article_texts.append(text)
return str(article_texts)
if __name__ == '__main__':
app.run('0.0.0.0')Let’s take a closer look at what has changed.
LIMIT on line 10 to limit how many articles we want to download.<p> (paragraph) tags is interesting content.parse_article function to get the actual content of the URLs that we found in the RSS feed and return that to the user instead of returning the URL directly. Note that we limit this to two articles by truncating our list to LIMIT for now, as the downloads take a while and Repl’s resources on free accounts are limited.If you run the code now, you should see output similar to that shown in the image below (you may need to hit refresh in the right pane). You can see text from the first article in the top-right pane now, and the text for the second article is further down the page. You’ll notice that out text extraction algorithm isn’t perfect and there’s still some extra text about “Share this” at the top that isn’t actually part of the article, but this is good enough for us to create word clouds from later.
Although Flask allows us to return Python str objects directly to our visitors, the raw result is ugly compared to how people are used to seeing web pages. To take advantage of HTML formatting and CSS styling, it’s better to define HTML templates, and use Flask’s template engine, jinja, to inject dynamic content into these. Before we get to creating image files from our text content, let’s set up a basic Flask template.
To use Flask’s templates, we need to set up a specific file structure. Press the “new folder” button (next to the “new file” button, on the left pane), and name the resulting new folder templates. This is a special name recognised by Flask, so make sure you get the spelling exactly correct.
Select the new folder and press the “new file” button to create a new file inside our templates folder. Call the file home.html. Note below how the home.html file is indented one level, showing that it is inside the folder. If yours is not, drag and drop it into the templates folder so that Flask can find it.
In the home.html file, add the following code, which is a mix between standard HTML and Jinja’s templating syntax to mix dynamic content into the HTML.
<html>
<body>
<h1>News Word Clouds</h1>
<p>Too busy to click on each news article to see what it's about? Below you can see all the articles from the BBC front page, displayed as word clouds. If you want to read more about any particular article, just click on the wordcloud to go to the original article</p>
{% for article in articles %}
<p>{{article}}</p>
{% endfor %}
</body>
</html>Jinja uses the specials characters {% and {{ (in opening and closing pairs) to show where dynamic content (e.g. variables calculated in our Python code) should be added and to define control structures. Here we loop through a list of articles and display each one in a set of <p> tags.
We’ll also need to tweak our Python code a bit to account for the template. In the main.py file, make the following changes.
from flask import render_template home() function to make a call to render_template instead of returning a str directly as follows.@app.route("/")
def home():
feed = feedparser.parse(BBC_FEED)
article_texts = []
for article in feed['entries'][:LIMIT]:
text = parse_article(article['link'])
article_texts.append(text)
return render_template('home.html', articles=article_texts)The render_template call tells Flask to prepare some HTML to return to the user by combining data from our Python code and the content in our home.html template. Here we pass article_texts to the renderer as articles, which matches the articles variable we loop through in home.html.
If everything went well, you should see different output now, which contains our header from the HTML and static first paragraph, followed by two paragraphs showing the same article content that we pulled before. If you don’t see the updated webpage, you may need to hit refresh in the right pane again.
Now it’s time to move on to generating the actual word clouds.
Once again, there’s a nifty Python library that can help us. This one will take in text and return word clouds as images and is called wordcloud.
Images are usually served as files living on your server or from an image host like imgur. Because we’ll be creating small, short-lived images dynamically from the text, we’ll simply keep them in memory instead of saving them anywhere permanently. In order to do this, we’ll have to mess around a bit with the Python io and base64 libraries, alongside our newly installed wordcloud library.
To import all the new libraries, we’ll be using to process images, modify the top of our main.py to look as follows.
import base64
import feedparser
import io
import requests
from bs4 import BeautifulSoup
from wordcloud import WordCloud
from flask import Flask
from flask import render_templateWe’ll be converting the text from each article into a separate word cloud, so it’ll be useful to have another helper function that can take text as input and produce the word cloud as output. We can use base64 to represent the images, which can then be displayed directly in our visitors’ web browsers.
Add the following function to the main.py file.
def get_wordcloud(text):
pil_img = WordCloud().generate(text=text).to_image()
img = io.BytesIO()
pil_img.save(img, "PNG")
img.seek(0)
img_b64 = base64.b64encode(img.getvalue()).decode()
return img_b64This is probably the hardest part of our project in terms of readability. Normally, we’d generate the word cloud using the wordcloud library and then save the resulting image to a file. However, because we don’t want to use our file system here, we’ll create a BytesIO Python object in memory instead and save the image directly to that. We’ll convert the resulting bytes to base64 in order to finally return them as part of our HTML response and show the image to our visitors.
In order to use this function, we’ll have to make some small tweaks to the rest of our code.
For our template, in the home.html file, change the for loop to read as follows.
{% for article in articles %}
<img src="data:image/png;base64,{{article}}">
{% endfor %}Now instead of displaying our article in <p> tags, we’ll put it inside an <img/> tag so that it can be displayed as an image. We also specify that it is formatted as a png and encoded as base64.
The last thing we need to do is modify our home() function to call the new get_wordcloud() function and to build and render an array of images instead of an array of text. Change the home() function to look as follows.
@app.route("/")
def home():
feed = feedparser.parse(BBC_FEED)
clouds = []
for article in feed['entries'][:LIMIT]:
text = parse_article(article['link'])
cloud = get_wordcloud(text)
clouds.append(cloud)
return render_template('home.html', articles=clouds)We made changes on lines 4, 8, 9, and 10, to change to a clouds array, populate that with images from our get_wordcloud() function, and return that in our render_template call.
If you restart the Repl and refresh the page, you should see something similar to the following. We can see the same content from the articles, however, we can now see the important keywords without having to read the entire article.
For a larger view, you can pop out the website in a new browser tab using the button in the top right of the Repl editor (indicated in red above).
The last thing we need to do is add some styling to make the page look a bit prettier and link the images to the original articles.
Our text looks a bit stark, and our images touch each other, which makes it hard to see that they are separate images. We’ll fix that up by adding a few lines of CSS and importing the Bootstrap framework.
Edit the home.html file to look as follows
<html>
<head>
<title>News in WordClouds | Home</title>
<link rel="stylesheet" href="https://stackpath.bootstrapcdn.com/bootstrap/3.4.1/css/bootstrap.min.css" integrity="sha384-HSMxcRTRxnN+Bdg0JdbxYKrThecOKuH5zCYotlSAcp1+c8xmyTe9GYg1l9a69psu" crossorigin="anonymous">
<style type="text/css">
body {padding: 20px;}
img{padding: 5px;}
</style>
</head>
<body>
<h1>News Word Clouds</h1>
<p>Too busy to click on each news article to see what it's about? Below you can see all the articles from the BBC front page, displayed as word clouds. If you want to read more about any particular article, just click on the wordcloud to go to the original article</p>
{% for article in articles %}
<a href="{{article.url}}"><img src="data:image/png;base64,{{article.image}}"></a>
{% endfor %}
</body>
</html>On line 3 we add a title, which is displayed in the browser tab. On line 4, we import Bootstrap, which has some nice CSS defaults right out the box (it’s probably a bit heavy-weight for our project as we have so little content and won’t use most of Bootstrap’s features, but it’s nice to have if you’re planning on extending the project.)
On lines 6-8, we add padding to the main body to stop the text going to close to the edges of the screen, and also add padding to our images to stop them touching each other.
On line 16, we use an <a> tag to add a link to our image. We also change the Jinja templates to {{article.url}} and {{article.image}} so that we can have images that link back to the original news article.
Now we need to tweak our backend code again to pass through the URL and image for each article, as the template currently doesn’t have access to the URL.
To easily keep track of pairs of URLs and images, we’ll add a basic Python helper class called Article. In the main.py file, add the following code before the function definitions.
class Article:
def __init__(self, url, image):
self.url = url
self.image = imageThis is a simple class with two attributes: url and image. We’ll store the original URL from the RSS feed in url and the final base64 wordcloud in image.
To use this class, modify the home() function to look as follows.
@app.route("/")
def home():
feed = feedparser.parse(BBC_FEED)
articles = []
for article in feed['entries'][:LIMIT]:
text = parse_article(article['link'])
cloud = get_wordcloud(text)
articles.append(Article(article['link'], cloud))
return render_template('home.html', articles=articles)We changed the name of our clouds list to articles, and populated it by initialising Article objects in the for loop and appending them to this list. We then pass across articles=articles instead of articles=clouds in the return statement so that the template can access our list of Article objects, which each contain the image and the URL of each article.
If you refresh the page again and expand the window using the pop out button, you’ll be able to click any of the images to go to the original article, allowing readers to view a brief summary of the day’s news or to read more details about any stories that catch their eye.
We’ve included several features in our web application, and looked at how to use RSS feeds and process and serve images directly in Python, but there are a lot more features we could add. For example:
If you’d like to keep working on the web application, simply head over to the Repl and fork it to continue your own version.
In the next tutorial, we’ll be looking at how to build our own Discord Chatbot. You can also find more tutorials like this one here.