According to the most recent studies, a recommender system is a type of artificial intelligence system that is used to recommend items to users based on their preferences. It uses data mining, machine learning algorithms, and natural language processing to analyze user data and make recommendations. 

They are used in a variety of contexts, from e-commerce to media and streaming platforms. For instance, in a streaming platform, such as Netflix or Spotify, they recommend movies or songs based on their users' past history. In the case of e-commerce, they recommend similar products to the ones that the user has already bought. 

The ultimate goal of a recommender system is to improve the user experience by providing personalized recommendations. How? By scanning large item catalogs for items the user might be interested in, and presenting them at the appropriate time. Let’s further explore the different kinds of recommendation algorithms, before discussing the benefits recommender systems bring to digital publishers.

Types of Recommendation Systems

Collaborative Filtering

Collaborative filtering is a type of recommendation system that uses the preferences of other users who have similar opinions and interests to generate recommendations. The main idea  is that users who have similar tastes and preferences will have similar opinions about the items they prefer. This will allow the system to make recommendations based on the opinions of other users in combination with the behavior of the individual

Collaborative filtering is used in a variety of contexts, from newsrooms to e-commerce, often being used to make personalized recommendations to users, such as articles or products they may like. This kind of recommender system can be implemented using various techniques, like user-based neighborhood methods, item-based neighbourhood methods or neural network based models.

User-based neighborhood methods look at the preferences of similar users to make recommendations.
Item-based neighborhood methods recommend items that many users have seen together with your preferences.
Neural network based models build a complex model that predicts which items you are most likely to prefer given your historic preferences..

Collaborative filtering allows for individual users to get a personalized experience using the data you already have on hand about your existing audience. By looking at similar profiles to the individual, recommendations go beyond simply what is most popular or trending. 

‍

Content Based Recommendations

Content-based recommender systems are a type of recommendation system that uses the attributes of items to make recommendations. They analyze the content of an item and match it with similar items to generate recommendations.

Content-based recommender systems use various techniques that include natural language processing and image processing. Natural language processing is used to analyze the content of items and generate features that are used to match similar items. Image processing is the use of a digital computer to process digital images through an algorithm. 

‍

Hybrid Recommender Systems 

A hybrid recommender system combines the best of different recommendation techniques to generate more accurate and personalized recommendations. Hybrid recommender systems use a variety of techniques to make a more comprehensive recommender system, such as combining Collaborative Filtering and Content-based Recommendations. 

‍

Context Aware Filtering

This technique is known to make better recommendations by using the context in which to give a recommendation. The context can for example be the type of device the user is using, or the article they are reading. You can find an in depth discussion of this type of recommendation system in our previous blog post.

‍

Direct Benefits to Editorial Teams

An important reason that editorial teams should be considering personalization is the impact it has on a user's experience with your digital channels. It has been proven that personalization leads to higher engagement which translates into users that are happier with their overall experience. 

Beyond this important benefit of using such techniques, editorial teams can also take advantage of increased content diversity, allowing editors to keep control, detecting fake news and the creation of filter bubbles, and the ability to gather more audience insights. We will explain each of these benefits in more detail. 

Diversity

Content diversity is a common worry within Editorial teams. Due to their wide catalog of articles, it’s difficult to make sure they show all of them to their audience, but more importantly, they show the right content that each reader wants to read. 

Thanks to recommender systems, users access less popular articles that wouldn't be shown if every user got the same editorially chosen articles or only popular recommendations. De Telegraaf found this benefit especially useful. They saw an overall increase in reader engagement and home page article diversity, all while serving relevant content throughout their homepage recommendations. Froomle recommendations increased diversity by 100% in terms of content topics shown compared to hand-picked selection. 

‍

Human-in-the-loop Approaches to Personalization

Another worry editors have when considering recommender systems, is the feeling of losing control over the user experience and which articles are being shown. However, we have designed solutions that enable the recommender system to be a support for the editorial team, rather than a replacement. This is what we call hybrid automation. It consists of a number of filters available to editors that keeps them in control of the content that is chosen, including topics, time since publishing etc. 

‍

Journalistic integrity: filter bubbles, fake news

Fake news and filter bubbles have been on peoples’ minds for a number of years and the rise of social media has only accelerated these concerns. Despite the continuous work towards early detection, it’s true that the negative impact of fake news can be immediate and affect thousands of people within minutes. However, the average user may not be aware of the effects of filter bubbles.

According to Oxford University, we define a filter bubble as a situation in which an internet user encounters only information and opinions that conform to and reinforce their own beliefs, caused by algorithms that personalize an individual’s online experience. Firstly referenced in 2011, many researchers have been trying to investigate more about this phenomenon and are trying to find smart solutions. 

Although there isn’t a clear solution for any of these problems, it’s been studied that recommender systems help in keeping journalistic integrity. Part of the research team at Froomle has been working on knowing more about filter bubbles. You can read our blog that further discusses filter bubbles here. 

‍

Audience Insights

Last but not least, a huge benefit for newsrooms is to get more detailed insights about their audience and how their content is performing. By using their first party data, they know their user’s past behavior and how they interact with different content. However, as a plus, editors can see not only their best performing content, but also content that was highly recommended. Knowing this information allows editors to pinpoint potentially interesting topics that may not have been as apparent previously. A great example is our recent report that used our research to show the difference in news consumption around the world. This data is invaluable to editors when deciding what and when to publish.

‍

Conclusion

The world of recommender systems can seem complicated at first glance. However, once the basics are understood, the benefits to editors are game changing. From increasing the diversity of content, allowing for hybrid control, helping to maintain journalistic integrity and providing valuable audience insights, recommender systems are a great partner to the newsroom. 

‍