The personalization of news sites is quite different from recommendations in other industries like e-commerce or marketplaces. For one thing, the ultimate goal in online publishing is not to sell but to engage: to keep users consuming content. As a result, user experience matters even more for news portals. In addition, with today’s 24-hour news cycle, loads of news items are published by the minute, only to lose relevance within a day or two – meaning a near-total daily turnover in the item catalog.
Interested in joining our staff?
Check out our open positions here.It takes the power of news personalization to harness this vast, ever-changing mass of information. At the same time, news editors strive to retain some control over recommendations in order to enforce consistent journalistic standards. Striking the right balance between automation and human oversight isn’t always straightforward.
So, how can news personalization cater to users’ interests while aligning with news organizations’ professional and business agendas? How is it best leveraged? What are the main pitfalls and thorny issues along the way? Read on to find out.
How News Personalization Makes a Difference for the Publishing Industry
In today’s attention economy, where money is being replaced by attention as our most scarce and valued resource, online news portals are in fierce competition for eyeballs. (And not just with each other. Social media, while being a vehicle for news to some extent, is also their number one competitor – in terms of both attention and revenue.)
Since content is so abundant these days that it has virtually no perceived value for consumers, most publishers resort to monetizing ad views as their primary source of income, while the subscription model remains a secondary option. As a result, these news organizations’ revenue depends heavily on visitor numbers and frequency, and their time spent on the site. Thus it’s crucial to keep people reading articles or watching news clips one after the other, and coming back for more of what interests them.
News personalization is the key to achieving this, and more. A powerful recommendation engine can complement the work of editors in story selection. Thanks to its computational capacity, it can elevate curating content to a level inaccessible by human effort only, while still leaving room for journalistic oversight. Relevant recommendations can keep readers glued to their screens. Paired with strategically timed prompts, they can even help to nudge the subscription rate higher.
For media consumers, on the other hand, the most pressing issue is information overload. They need to be able to focus on what’s timely, relevant, and important, and not drown in distractions. News personalization can be tremendously useful for them in streamlining their reading list to a manageable size, and in separating signal from noise. Which, in turn, helps publishers, too, by increasing reader loyalty.
On the flipside, as filtering functions users rely on often unwittingly, recommendation systems (or rather, the newsrooms employing them) have a huge responsibility in shaping people’s perception of the world.
Measuring the Performance of News Personalization
It follows from the above that the key performance indicators (KPIs) of news personalization systems primarily assess user engagement resulting from recommendations. The most frequently used metrics are
- Time spent on site – the duration, measured in seconds, of an individual user’s browsing session on the news site. This KPI helps to gauge the site’s overall performance as well as user engagement.
- Number of article pages viewed from recommendations – counting the article pages opened by clicking on recommendations, this metric also helps to qualify the impact of recommendations on user engagement.
- Click-through rate (CTR) – the percentage of recommendations that a user clicks on.
- Conversion into views (long-term CTR) – as opposed to CTR, this metric indicates the percentage of recommendations that garner not just clicks but actual content views. We consider news content to be viewed if the user has read or watched at least ten percent of it. Alternatively, a tracking code can be placed at the bottom of an article page to signal if the user has scrolled down to the end of the article, in which case we assume she has read it all. Conversion being a narrower definition of engagement, it helps to filter out clickbait content and unintentional clicks, to focus on the performance of quality journalism.
News Personalization Use Cases
When it comes to media consumers’ time spent on the news site, every aspect of the user experience can have an impact, from the general look and feel to the structure of the website or the arrangement of text and content on its pages. Another key aspect is news personalization, which comes in many shapes and forms, the most basic being article recommendations.
Recommended Content
Recommendation widgets display tailored recommendations for the reader on the main page of the news portal, as well as on section and article pages. By default, these can appear as a personalized list of titles or as a slider of content recommendation boxes.
On the article pages, a more contemporary format of news personalization echoes the news feed feature of social media platforms. As soon as the reader has reached the end of an article, the next recommended piece appears below, so all she has to do is scroll down to continue reading the most relevant content. By minimizing the number of clicks needed to keep readers on the site, this function, called infinite scrolling, has proven useful for Yusp client La Vanguardia, one of Spain’s leading news outlets. Executed well, infinite scrolling makes for a seamless, immersive reading experience, especially on a tablet.
The choice of relevant recommendations in a given moment depends largely on the way the reader consumes content: whether he’s skimming headlines and nipping in and out of articles to get the facts as quickly as possible, or taking the time to dive deep into a topic. Because differences in media consumption are usually linked to situation and context, data such as day of the week, time of day, or device used is also taken into consideration.
As a result, news personalization engines tend to recommend breaking-news type articles on mobile on a weekday morning, and feature more long reads or background pieces when the news site is accessed from desktop on a weekend afternoon. In addition, geolocation data can also be leveraged to select stories that are more relevant in the user’s region.
Tailored Updates
Customizing newsletters and push notifications is a widely used form of news personalization, because there are a number of hurdles that can be cleared with its help.
To begin with: the users of news sites are not particularly compelled to register – as opposed to ecommerce, where signing up is generally considered part of the purchasing process. Therefore, the first challenge is to build a newsletter database.
Overlay signup forms can do the trick, but the timing is crucial: they work best if displayed at a moment when the reader’s engagement is high – that is, when she is truly enjoying the content offered to her. Personalization engines can detect these points in the user journey, when a reader has clicked on and fully read recommended articles. So they can pounce, and prompt her to sign up for the newsletter when she is most inclined to do so.
Once users are on board with getting content from the news site delivered to their inbox, the next challenge for news personalization is to get them to actually open the newsletter. To this end, the subject header of the email has to be customized, as well as its content. A common tactic is to feature the title of the article the addressee is most likely to find compelling.
The open rate of newsletters is tracked to see how well they perform. Reading newsletters signals a degree of user commitment that can possibly be taken to the next level, subscription.
Push notifications being a more instant, but also more intrusive form of updates, they are best used for delivering breaking news at a moment’s notice. There’s less room for personalization; instead, users are generally encouraged to choose which topics they want notifications about.
Cross-Platform Recommendations
News personalization can also be used to boost traffic between partner publishing sites. This is particularly helpful when a news portal launches a new affiliate. Sampling its articles in the main outlet’s recommendations can help overcome the sister site’s “cold start problem”. Such a setup needs prior agreement from all publishers involved, as the personalization provider stores their user data in separate silos.
Personalizing Ads
Because the majority of news portals live off ad impressions, the choice of ads or paid content accompanying articles and videos is crucial for their success. Yet finding relevant (and transparent) advertising that truly complements editorial content and provides a seamless user journey across media and ecommerce platforms has always been a hard nut to crack.
Recently, there have been initiatives to involve news personalization engines in solving this problem. Yusp is currently experimenting with a system that pairs news articles with relevant offers by the publisher’s ecommerce partners or third-party advertisers. In turn, these content-ad combinations are matched with potentially interested readers.
An alternative solution is native advertising (offered by Taboola, for instance): serving a selection of personalized ads alongside article recommendations. Native advertising providers offer publishers a share of the ad revenue, which makes them financially more attractive than news personalization, a paid service. On the other hand, an evident drawback of native ads is that they direct traffic away from the news portal, undermining long-term user engagement.
News Personalization Challenges and Best Practices
In the news industry, personalization faces a unique set of challenges, such as integrating editorial control, or finding the right approach to manage clickbait content. As for the issues the news industry itself is struggling with, like making the subscription model viable, recommendation systems can be a useful tool for tackling them. News personalization providers have developed best practices for either type of challenge. Let’s look at these in more detail.
Setting the Degree of Editorial Control
As mentioned earlier, news personalization shapes people’s media diet and thus, their outlook on the world. As such, it cannot be left entirely to algorithms. On the other hand, the sheer mass of news content published every minute makes recommendation systems indispensable. To what extent, then, should humans stay involved in selecting articles to recommend?
The editors of news sites aim to retain some control over personalization. This is to make sure that leading articles are included in recommendations across the board; or that content requiring a lot of professional resources, such as investigative journalism, does not end up buried under trending tabloid pieces. To serve newsrooms best, providers like Yusp offer a range of personalization options, from full-on “autopilot” mode to various combinations of manual and algorithm-powered content selection. Here are two specific examples:
- Selecting key stories to show all readers on the main page is a task editors typically prefer to do themselves, as a means of setting the agenda for public discourse. However, the degree to which specific articles on this shortlist are displayed prominently to individual users can be finetuned by the recommender system.
- As a simpler alternative, an algorithm can track which key articles a user has already read, and choose another story from the “must-see” list to dominate the main page next time the same visitor returns.
In close collaboration with their clients, news personalization systems help to ensure that editors can keep control over recommendations where it matters most to them. By harnessing the power of automated recommendations in other parts of the news site, newsroom staff can focus on what they do best: producing quality journalism.
Driving CTR or Supporting Conversion
If left to their own devices, people tend to gravitate towards popular content: celebrity gossip, cat videos, and clickbait headlines. This poses a dilemma for news sites: while such “easy reading” material provides instant gratification to readers and more revenue from CTR, it may erode user engagement and loyalty in the long run.
Faced with a dilemma like this, the choice of KPI makes all the difference. If CTR is a news outlet’s North Star metric, then creating easy access to the most in-demand topics, however shallow, is in its best interest. If conversion and overall user experience matter more, then it pays off to drive traffic to quality content that may be overlooked otherwise. There are industry trends pointing in both directions, corresponding to the tabloid-broadsheet faultline in print media.
Whichever business decision is taken, news personalization engines can help to filter and prioritize content accordingly. To drive CTR, the recommendations will focus on “Most Viewed” topics. If the goal is to give more emphasis to quality journalism, the system will consider other selection factors beside popularity, like how long an article was viewed or what percentage of it was read. This way, the algorithm can recognize if many people were disappointed by a clickbait piece, and rank it lower or remove it from the recommendations altogether.
Growing the Subscriber Base
Converting casual readers into subscribers is a major challenge across the news industry. Though personalization is not the most important means of achieving this, it can help to move the needle.
With the introduction of the paywall, today, news outlets offer various combinations of free and paid content. One possible model is allocating each reader a limited number of free articles, then, upon free registration, making another fixed number of articles available to them, before they are required to subscribe in order to keep reading.
Because paid content is not built around ads, the business objective here is conversion, not CTR. High quality, highly relevant articles are needed to convince readers that paying a regular fee, in return for access to unlimited content, is well worth the investment. A news personalization engine can distinguish between free and paid content, and include in a free reader’s recommendations “sneak peeks” at paywalled articles that are intriguing enough to prompt subscription.
Behind the paywall, the role of personalization gets more emphasis. Subscribers need to feel constantly justified in their choice for paid content, and this can be achieved with recommendations showcasing the best, most timely, and most relevant articles. Because subscribers provide more information about themselves than non-registered users, personalization is also likely to be more efficient behind the paywall. This can be interpreted as a positive feedback loop: the more committed readers are, the better their user experience will be.
Overcoming the Recency Issue and the Cold Start Problem
Due to the ephemeral nature of news, the item catalog of news sites is in constant flux. Articles that are “hot off the press” in the morning are obsolete by the afternoon. As a result, news portals need to churn out a tremendous amount of fresh content every hour. In order to keep recommendations up to date, instant syncing with the news personalization engine is imperative.
On top of this, news outlets face a challenge called the recency issue: the more recent an article, the more relevant it would be to a large audience – if only they could discover it. The average news piece gets 80 percent of all its clicks within 24 hours of going live. But to make the most of this initial period, and direct traffic to brand new articles with practically zero interactions is a bit of a challenge for recommender systems.
The first few readers are the “guinea pigs” of the personalization process: their interactions inform the collaborative filtering algorithm as to who else might be interested in the article. All told, it takes about a hundred clicks on a piece of content to make it recommendable, but time is of the essence. Yusp is outstanding in that it’s able to retrain its machine learning models as often as every few minutes, leveraging new user data almost instantly. This greatly reduces the cold start phase of news pieces.
Of course, there are sections and topics where articles remain newsworthy for a longer period. News personalization systems adapt by considering section or category information as well as publication date when ranking articles to recommend. Yusp used this section-based filtering for its client La Vanguardia. Content in the Politics and International news sections could be recommended for no more than 24 hours after its publication, while articles in the Tech and Culture categories stayed recommendable for up to a week.
Some news content may regain relevance seasonally, or when certain events occur. Tracking changes in CTR can help to pinpoint instances when demand for an article spikes anew. This way, recommender systems can facilitate the republishing of existing material, enabling newsroom staff to cut some corners.
Breaking Out of the Filter Bubble
Filter bubbles, formed by recommendations that cater to a user’s views and preferences so much as to become a sort of “echo chamber”, have been cause for concern lately. However, these are primarily an issue in social media news feeds that aggregate content from various (often obscure) sources. News outlets are typically aligned with a political direction and prioritize certain topics over others – some may focus on the environment, others on social justice, or lifestyle content. These characteristics function as preset filters that readers are aware of, or even seek out, when choosing a specific news portal.
To further minimize the risk of “over-personalization”, news organizations and their personalization providers must work together to ensure a healthy balance between tailoring a reader’s media diet and staying somewhat random. Recommender systems can help by serving a certain amount of serendipitous content – articles, topics outside a user’s known circle of interest – that enables discovery and thus enlarging this circle. By applying editorial control over recommendations, newsroom staff can make sure that essential pieces form a sort of common ground for all readers, so they’re not entirely cut off from each other in over-personalized content bubbles.
Tap Into the Yusp News Personalization Expertise
If you’ve read this far, it’s safe to assume that news personalization is highly relevant to you. To learn more, read all about our solutions for the leading Spanish news outlet La Vanguardia. Immerse yourself in our Deep Learning explainer to understand how this function can be leveraged for news personalization.
Got specific questions? Don’t hesitate to get in touch, and let’s see what we can do for your news organization.
