Recommendations Part 1: The need for enhanced audience behavior insight
How we think recommendation systems can help us increase engagement and revenue
In order to ensure the Local News Lab builds the right product to solve the right problem for the right newsrooms, we do a lot of research, both academic and hands-on interviewing of newsroom leaders and staff.
Much of our work to date has centered around what news organizations need and how we can help them optimize their revenue strategies by using machine learning and AI-informed products. However, a big piece of the puzzle is the audience. But audience members aren’t as easy for us to reach as the folks publishing the news, nor are they a monolith. It is for this reason that we began to discuss reader habits: if we could learn more about how readers behave on a news site, we might better understand what makes them tick (and click). This led us to the idea of recommender systems.
What is a recommendation system?
A question that has come up frequently in our interviews with newsroom leaders is how to properly segment readers in order to more finely target various campaigns for engagement such as subscriptions, donations or newsletter sign-ups. Recommender systems are able to easily distinguish between different “kinds” of readers based on their interests and reading habits. This is made possible by an algorithm that analyzes how a reader travels through the site and compares them to other readers. By analyzing the correlations between similar readers’ behaviors, the algorithm is able to predict what they are likely to want to read next. These predictions can then be displayed as a “Read Next” module on the news website to encourage readers to engage more deeply.
This dovetails nicely with the business objective of moving readers along the funnel, something which the right recommender system can be optimized to do by recommending specific actions or high-value content.
What problem will this solve?
A recommendation system can help filter down the entirety of the content available for readers, and connect them more quickly to content that is more likely to be engaging to them personally (thus potentially deepening their relationship with the brand). It can draw more value from evergreen content and enable strategies like serving fewer, but more impactful calls to action (for donations, subscriptions or registration sign-ups) to individual readers.
How can it help newsrooms increase engagement and revenue?
A recommendation system can provide the scaffolding for all kinds of revenue-oriented content strategies — personalized newsletters/reading lists, smart paywalls, auto-content-locking, and more.
This comes part-and-parcel with a machine-understandable representation of readers and articles according to interests and level of engagement. In simpler terms, the recommendation system condenses the entirety of a reader’s history into a small set of explanatory numbers which represent the reader’s interest level in various topics that can be then used to predict their next click, dwell time, likelihood to churn, or likelihood to subscribe. These predictions can help us create smarter and more personalized calls to action which should then (we hypothesize) lead to increased engagement and revenue. In order to test these hypotheses, we came up with an experiment: a recommendation module to be placed on article pages of a newsroom partner’s website.
This experiment is about to launch and we will post a sequel to this article in the coming weeks with a technical deep dive into how we built our recommender system and intend to evaluate it.
