Everything ought to be smart today: from smartphones to smart homes, smart grids and smart medicine … the list goes on. What they all have in common is the idea that products or services should not be one-size-fits-all, but rather adaptive to the user.
Our smartphones do this via the myriad of available apps, as do smart TVs. Smart cars will be autonomous and entertain us while driving, whereas smart medicine promises better diagnosis and therapy.
In short, they all personalise the user experience.
In 2016, at Neue Zürcher Zeitung (NZZ), Switzerland’s newspaper of record, we were lucky to receive a Google Digital News Initiative grant to build a prototype of a truly smart, personalised news experience: the NZZ News Companion.
In news media, more often than not we are still thinking in terms of one-size-fits-all products. This is suboptimal. Personalisation, done right, is actually in the best interest of the reader.
The use of news landing pages is a prime example. Landing pages are highly volatile, with a rapidly changing selection of featured articles on the landing page throughout the day. This means readers can miss out on a lot of interesting stories — especially the ones that might interest them most.
Pair this with the observation that most readers don’t bother to browse beyond the landing page, and it is obvious that what is somehow considered by most readers to be a selection of most relevant articles actually is nothing but a snapshot.
I am not suggesting that we should decrease volatility of landing pages. This volatility is just a symptom of a deeper problem: that computer screens and mobile screens provide very limited space — and certainly not enough for all the news.
We seriously need to think about engineering a more user-tailored, smart news experience.
For us, the NZZ News Companion is the archetypal smart media product, providing everyone with their own smart, adaptive, and context-aware news stream. Perfect information. No unnecessary distraction, but also no filter bubble. The result is a news stream that is highly relevant and upholds journalistic standards.
The most important part when it comes to data products and personalisation in particular is the underlying algorithm (i.e. how we chose to compute the article recommendations). These days when people talk about algorithms, what they mostly mean are algorithms for statistical inference, which detect patterns from data. However, these methods can be somewhat blackboxy, in the sense that you put data in and get some results out without knowing why.
Considering our goal to build a news stream upholding journalistic standards, this is somewhat discomforting. More critically, machine learning algorithms are not designed to shape the future; they are designed to replicate the past. This makes them more vulnerable to create effects like the filter bubble.
This is why, in the case of the NZZ News Companion, we first spent much more time on algorithm design, which means understanding the desired user experience. Only then, we decided about what methods/algorithms can achieve this in an automated way. In our case we envisioned the following user experience:
- Readers should be recommended articles for up to two days if the articles remain relevant.
- Readers should see articles that match their interest early after publication and high up in their recommendations.
- Readers can’t miss relevant news, even if it is not in their “proven realm of interest.”
- Readers should always feel well-informed and never fear being caged in a filter bubble.
Boiling all of this down, the challenge translates into finding good measures for two principle concepts: general relevance and personal relevance. Once these are quantified, we put them together with some additional business logic into a recommendation algorithm to create the NZZ News Companion.
We found that general relevance — the idea that some content is valuable for every reader — is hard to define algorithmically. How is a machine supposed to know what kind of content might be relevant, given current and past context? While it is generally possible to create such a decision algorithm, it’s just impractical at this point in time, especially given the fact that we have journalists creating our stories with exactly this knowledge of relevance in mind.
So we opted to create a metric for general relevance that we called “editorial relevance.” It’s based on quantifying how articles have been placed on our Web site by our editors, and subsequently deduces relevance from placement. Simplified, one could say the more prominent and the longer a certain article is on our Web site, the more editorial relevance we give that article. We found this to be a rather simple, yet powerful concept.
Personal relevance is not less tricky, but at least we can use some machine learning here. The idea is simple. We observe what you have shown interest in previously (which stories you have read) and learn your interests. If you are always reading news from “finance” in the morning and “international” in the evening, we can pick up that pattern and adjust your recommendations accordingly.
This can become much more complex when you take more features into account, like the author, length of an article, tags, and more. Also, using some advanced natural language processing techniques, we are able to closer match interests, beyond pure meta information. This allows us to not only recommend articles that match user interest on a high level, but also in a more fine-grained fashion.
We are still at the very beginning of our journey to a truly smart news experience, and we will continue experimenting and iterating over the next months and years. Most importantly, we need to test and improve these techniques “in the wild.”
One particular insight we learned already is that having one news stream that is too adaptive is somewhat counterintuitive for many readers. There needs to be some kind of predictability and clear communication why certain stories are seen, and this is as much a communication challenge as it is a technical one.