For the last decade, personalisation of content has been associated with filter bubbles. Initiated by Eli Pariser’s theory in his book The Filter Bubble: What the Internet is Hiding From You, the assumption that algorithms would create an environment in which you only encounter familiar opinions or information has spread like wildfire. Only a few studies have profoundly analysed the impact of relying on algorithmic personalisation for surfacing content on online newspapers.

Based on our experience in news personalisation for various digital newspapers around the world, we put forward several avenues that tend to demonstrate that personalisation is more of a solution than a problem in relation to this phenomenon.

1. Personalisation for news Web sites can be modular

News Web sites, apps, and other digital channels are frequently made up of “blocks” of content, which serve different purposes, such as breaking news, highlights from a specific section, and recent news. Therefore, it is very common to only apply algorithmic personalisation to specific blocks.

This personalisation engine can power each block with a different treatment, such as by focusing on distinct objectives. This modularity is key to ensuring a diverse distribution of content and, therefore, deconstructing filter bubbles by design. This is unlike platforms like Facebook, LinkedIn, or Google that rely solely on one content feed.

2. Personalisation increases diversity of content

Deciding which content to present on the different pages of your Web site can become a very tedious task given the importance of keeping that content fresh and updated within a small window of time. When relying on a manual selection of articles, it takes time to select the right content and input this into the system.

Personalisation, on the other hand, automates this task in real time and makes it possible to update the content selection in milliseconds for every individual user visiting your Web site. As a result, it has been proven that personalisation can recommend significantly more articles than a human (team) can on a daily basis.

A great example comes from De Telegraaf, the leading digital newspaper in the Netherlands. With manual selection, it recommended around 250 different articles per day while the personalisation solution shows more than 600 different articles.

Personalisation also does a great job of balancing articles from different sections. Manual selection tends to focus predominantly on general news articles while personalisation proposes articles from a wide range of categories.

3. Personalisation fights human bias, helping the newsroom define clear editorial guidelines

Another less visible advantage of deploying an algorithmic solution for content curation lies in the organisational process to achieve it. One part of that process inevitably requires the editorial team to define clear rules for the algorithm to follow that determine which articles should be promoted or blacklisted, the required freshness of content for different positions on the Web sites, the ratio of paid versus free articles, and more.

While you can rely on the algorithms to solve most of these questions, it is often best to discuss these parameters as a team before launching the project to ensure a maximum buy-in.

This creates a real opportunity for the team to structurally fight against filter bubbles with clear editorial guidelines.

4. Personalisation can use reader behaviour to help navigate from one topic to another

There is a myth surrounding the process of computing recommendations for a reader: In other words, proposing “relevant” content would mean promoting content similar to what you’ve already read, hence locking you into a limited amount of content.

The reality is quite different. Recent algorithms have dramatically improved the quality of recommendations and their ability to take you from one content piece to another. More specifically, the replacement of content-based approaches (looking at the characteristics of content you liked, such as the topic, and recommending more of that) for behavioural approaches (looking at what readers usually read) has played an important role in breaking these filter bubbles.

Since then, personalisation systems that are anchored in real-time computation of your preferences outperform humans in proposing a diverse and balanced content selection that supports readers in their discovery of your content.

Going forward

What does this mean for you and your personalisation strategy? It all comes down to implementing the right strategy in the right way.

As we’ve seen, a modular approach identifying different objectives for each content block will unearth diverse topics and increase readership. Automated personalisation is also significantly more diverse than manually selecting and provides a better balance of topics.

Personalisation can also help with the newsroom by improving consistency of your approach and buy-in across the organisation. Lastly, personalisation takes readers on a journey from one topic to the next, proving that readers have more than just a single interest and the right personalisation strategy supports their discovery.

. . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . .