Machine Learning and Artificial Intelligence for Content Publishers
The Robots Are Here, And It’s Okay
Over the last few years, machine learning (often abbreviated as “ML”) and artificial intelligence (“AI”) have quickly evolved from science fiction fantasies to real solutions for challenging problems. Though they’re best known as the tech behind self-driving cars and facial recognition, AI and ML represent a powerful toolset that stands ready to tackle a whole host of opportunities for content authors and publishers — from big-name publishing brands to those just getting started.
Even more importantly, AI and ML are a chance for forward-thinking publishers of all sizes to get ahead of their competition. The technologies are still relatively new, and not widely adopted. They empower significant improvements in the editorial process, and in user engagement: adding a new avenue for growth, whether your content is your primary product or a secondary supporter for marketing and storytelling endeavors.
DDI Editor's Pick: 5 Machine Learning Books That Turn You from Novice to Expert - Data Driven…
The booming growth in the Machine Learning industry has brought renewed interest in people about Artificial…
To best understand the opportunities that these technologies unlock, let’s begin by learning more about what they are, and how they work.
What are ML and AI?
In our science fiction fantasies, artificial intelligence is taken to the full breadth of the term: a computer-generated version of the human brain, capable of understanding, analyzing, learning, and acting, just like we are. Though we’re not there yet, the modern state of artificial intelligence has largely grown from the realization that computers are excellent at two fundamental tasks in which the human brain also excels: recognizing patterns and making educated guesses based on those patterns.
Computers, however, are capable of working with a far broader set of data than our brains are. We’re able to compare three or four variables across a few dozen examples to recognize a trend. Computers can tackle thousands of times of that.
The art of modern artificial intelligence is in recognizing how problems can be decomposed into observations that computers are well-tuned to make, and building the approach, often called a “model”, that lets them do so most effectively. We can even take this a step farther, by building models that are able to retrospect on the effectiveness of their guesses and adjust their future predictions based on the accuracy of their prior ones — we call these models “self-learning.”
Artificial intelligence has been appearing across a number of industries over the last several years. Perhaps the most commonplace example is in real estate price prediction, popularized by Zillow, Redfin, and other sites. Fundamentally, real estate price comparison is driven by analyzing the similarity of properties across a number of facets and extrapolating a given property’s value based on how it stacks up. Historically, this was often done with manual analysis. This works well when we’re comparing across a small handful of variables, like home size or the number of bedrooms.
As we begin to consider many more variables, like lot size, features, location, time since the last renovation, home age, and more, this becomes significantly more complicated. Zillow and Redfin realized that this type of data is tailor-made towards the strengths of computers, and have built models that are able to analyze this data and continuously adjust their conclusions as the market continues to evolve.
Machine learning is a technology that empowers artificial intelligence. Though I’ll spare you the details, you can think of it as the myriad approaches, built atop the field of statistical analysis, that enable us to develop models for artificially intelligent conclusions.
Though the most obvious use-cases for ML and AI are numerically-based, it’s not a requirement: there’s been a lot of research done over the last few years to develop efficient ways to turn images, speech, text, and other types of data into forms that ML and AI are able to effectively work with.
Now that we know what ML and AI are, let’s explore a few examples of the cool things that they can do for content publishers today. Though you’ll ultimately need to qualify the benefits of each within the context of your specific business, the vast majority of publishers I’ve encountered during my work at 10up are always striving to increase user engagement, content production productivity, and content efficacy: all things that ML and AI are particularly good at.
Most content publishing websites are divided into a handful of unique top-level categories, often with further subdivisions underneath. For sophisticated publishers, this schema is typically defined during an Information Architecture exercise during the website’s initial construction, leveraging insights from the publisher’s editorial team surrounding the content that they intend to write.
Over time, even the best-planned content architectures fall victim to natural shifts in the publishing environment and ambiguity in the true classification of content that sits “in between” categories. Using artificial intelligence solutions developed by IBM Watson, along with ready-made solutions like 10up’s ClassifAI plugin for WordPress, we’re able to holistically re-examine content to better understand what it’s truly about. Beyond providing insights when considering re-classifying content, this approach can be extended for automatically migrating content into a new content architecture.
There’s also an ancillary benefit: these technologies are very similar to how search engines read and understand your content. In reviewing the outcomes, you may surface new trends and patterns that are otherwise not directly evident.
We can take the notion of classifying content one step farther and aim to develop insights into why content performs the way that it does. What content does your readership most connect with? What topics are more likely to go viral? Are particular authors, particular publishing times, or specific channels more likely to lead to content that’s loved by your readership?
Though some analytics platforms, like Parse.ly, are beginning to think about the insights and decisions that can be drawn from this data, my mind imagines an editorial planning tool that illuminates the entire editorial process in this light. Some publishers, like Buzzfeed, have leveraged similar approaches for years with great success — and we’re now at a point where the technology is available for everyone to do it, too.
High-quality content recommendations are the singularly most effective tool for retaining readers and converting one-time users into a dedicated audience member. Historically, content recommendations are driven by the popularity of content categorization and popularity, building off the notion that a reader is most likely to be interested in content about similar topics that other readers have also enjoyed, too.
While this conclusion isn’t incorrect, it only scratches the surface of what’s possible. E-commerce companies like Amazon have built massive businesses over multi-faceted recommendations, understanding that no two users are alike, nor are most user’s interests one-dimensional. Leveraging thoughtfully-crafted machine learning models, it’s now possible to build content recommendation engines that utilize reader’s historical behaviors, and in-depth understanding of the nature of the content, to anticipate the content that a specific reader is most likely to engage with. When these recommendations are positioned well, particularly on mobile devices, they’re a sure-fire way to transform your content into a black hole that sucks readers in.
We can even apply this same technology to other facets of our content platforms, like search — delivering recommendations that are specifically tailored to each unique user, backed by the knowledge acquired by analyzing the user base’s behavior as a whole.
It’s also possible to leverage AI and ML to tackle the very heart of the content publishing process: generating content. Though it’s unlikely to ever replace thoughtful, nuanced journalism, AI has already been successfully utilized to capture rote content production, like post-round performance summaries from the PGA Tour. Over time, I expect this trend to continue to expand, with revenue generated by routine content used to fund insightful original reporting, much like Buzzfeed has begun to do.
Getting Started with AI and ML
The world of machine learning and artificial intelligence holds a great deal of promise, but the horizons it is forming can make it challenging to grasp the full breadth of its potential and differentiate time-wasting features from those that hold real promise.
Over the next few years, I expect a number of companies to emerge from those that provide productized solutions to more commonplace problems. These “features” that don’t require a fundamental reimagination of how your content business works are a great place to begin building familiarity and trust in the promise of AI.
When you’re ready to take a step farther, your best bet is finding an experienced technology strategist to help you unpack the core challenges specific to your content and develop AI-backed solutions that lean into those challenges. Machine learning and artificial intelligence represent a complete reinvention of the challenges we can solve and the solutions we can develop, and there’s no harm in getting help along your journey.
Over the next few weeks, I’ll be exploring examples of artificially intelligent solutions for several other common challenges faced by content publishers. Be sure to sign up below to be notified as they’re available!
Originally published at Phil Crumm.