Improve Content Discovery with Emotionally Intelligent Recommendations
As massive amounts of content are published online every day, and brands look to boost engagement with their visitors, a familiar scenario around content discovery arises: How to offer interesting content to individual users? How to know what’s interesting to that person?
Solution? Make content recommendations emotionally intelligent.
If you clicked on a title like ‘emotionally intelligent content discovery’, I’m assuming you know something about content discovery platforms, but let’s start with an overview of how recommendations systems work, and then look at how to improve the quality of the metadata to power content-based recommendations.
Recommended for you
Recommendations are one of the most effective techniques for content discovery. You’ve seen this on pretty much every e-commerce and media site:
Brands benefit by offering relevant suggestions, recommending things ‘ you might like’ to keep visitors engaged with their content or sell more products. If the recommendations are valuable to visitors, they also benefit by discovering new content they may not have found otherwise.
Good recommendations will help build happy and loyal audiences. Bad recommendations, however, can turn users off, so it’s increasingly important to design a recommendation system and metadata strategy that goes beyond traditional targeting approaches.
Building a Recommendation Engine
There are two main approaches to designing recommendation engine, content-based and collaborative filtering, frequently used together. Custom algorithms leverage internal and external data to suggest related content/products, link users with similar interests, forecast user actions based on past behavior, and ultimately surface new, relevant content that leads to an action or behavior (purchase, click, view, share, spend time on site, etc.)
Internal data is gathered from the site and content itself, referencing metadata added to content and page to make connections between content. Content recommenders can also take into account user past behavior and profile actions taken. External data can be gathered from trending topics on social media or Google Trends, as well as collected in customer surveys or sentiment analysis, and provides recommendation engines information about relevant content topics that are ‘popular’ and ‘trending now’ to promote.
Tagging is a common concept that allows publishers to attach specific metadata to content, often as an additional field in the CMS, that describe attributes of that particular piece of content.
Recommendation engines reference those tags and other shared categories (genres, content type, etc.) to link content together, and determine what to surface/recommend when a user visits page A.
By analyzing characteristics or attributes of content (historical items viewed, read, listened, shared, liked, upvoted, rated, etc.) it’s also possible to get an idea of what content visitors prefer, and predict what other content would appeal to the same audiences.
Collaborative filtering references the wisdom of the crowd in order to provide recommendations. Think of ‘most popular’ or ‘trending now’ content, and of Amazon’s product pages that use sophisticated algorithms to generate “Customers Who Bought This Item Also Bought…” and “Frequently Bought Together” sections.
Collab filtering is largely powered by correlations — if a user likes item X, Y, and Z and another user in the community also like X, Y, Z, and I, the chances are the former will also like I.
Another example is Mashable, who has an in-house tool called Velocity that aims to predict how popular stories will become on social networks, using detailed analytics about how stories are shared and interacted with across social networks. Featured in homepage categories and on article pages themselves, this added transparency into what makes something trending helps build trust with users, and helps the Mashable editorial team find leads and larger trends.
Ephemeral recommendations are a third approach that takes into account only the current action to suggest relevant content, trying to catch you on impulse. This is frequently used in ecommerce, recommending products related to what you’re looking at right now, or items you’ve added to your cart. It’s also used in ‘cold start’ scenarios, when not enough data about a user is available to provide more personalized recommendations.
All content recommendation systems depend on:
- What is known about individual users (which can lead to concerns about privacy or be limited by industry rules)
- What is known about individual pieces of content (products, stories, videos, photos, etc.) — how are they tagged in the system?
- Information about the audience as a whole (demographics, behavior)
- Domain specific information sources (trends, tribal knowledge)
While valuable tools in content discovery, there are limitations of recommendation systems that can have a negative impact on a brand, if not properly monitored.
For instance, if recommendations are too pushy, people will learn to ignore them. If recommendations are not useful or relevant, people resent that the brand doesn’t know them. A lack of explanation on why recommendations are made can exacerbate someone’s lack of trust.
Other potential downfalls come from being too reliant on algorithms — for instance, recommending the same ‘most popular’ content to all visitors will only increase and reinforce the popularity of that content, and bias the system from promoting other relevant content to a specific audience.
By focusing too much on big data targeting or marking initiatives, recommendation systems are blind to emotion and what is truly the best experience for users.
How can we make recommendations more emotionally intelligent?
The process to make more accurate content-based recommendations involves identifying the emotional attributes of content, to create a more robust tagging strategy.
A simple framework, proposed by content strategist Michael Andrews, helps to classify content based on content experiences rather than content topics, which adds an another dimension to content-based filtering.
- First, identify general interest content — content that audiences might find interesting even if they weren’t searching for it specifically.
2. Then, identify qualities of your content and tag your content. Go beyond your brand voice / style guide, and think about:
- What’s most distinctive about your content?
- What do audiences most relate to?
3. Set up your recommendation engine.
4. Monitor performance, and adjust.
Front-loading the work by defining a tagging strategy and adding useful tags to your content will make it trivial for a recommendation engine to churn out relevant suggestions.
Better metadata = better recommendations
Let’s dive deeper into how to use metadata tags to describe your content experience, and how to characterize content according to multiple, distinctive qualities, a concept referred to as content attractors.
“A content attractor is a quality of your content that resonates with certain audiences. It may be your approach to talking about a topic, or your point of view. It produces an emotional experience. Often it is the combination of two or three qualities that makes content distinctive and special.” — Michael Andrews
Here are some tips to create meaningful, emotionally intelligent metadata:
- Think in terms of adjectives and emotions, and tag content according to its emotional qualities. For example, surface stories based on a reader’s mood by tagging the tone of the story (e.g. uplifting, serious).
- Reduce complexity by identifying similarities — focus on general interest content, and don’t be so specific that there will only be a few pieces of content that relate.
- Ask visitors what they like/don’t like most about certain content — listen for when someone mentions dimensions such as the style of the content, its perspective or point of view, its approach to help, and the kind of occasion it would be viewed.
- Go beyond content types and topics — beyond taxonomies that focus on concrete attributes (nouns), look to typologies that focus on qualities (adjectives and concepts) to make relationships between content.
The following questions can help define what makes your content distinctive and attractive to your audience.
Does your content have a distinct attitude?
Define elements that makes content distinctive, e.g.:
- authoritative — access to the most reliable information
- exclusive — preview privileged info
- trust our picks — we’ve found the best for you
- contrarian — don’t rely on conventional wisdom
- approachable — we make the difficult approachable
- visionary — show how the future will be different
- championing, crusading
- practical — only the stuff you can use
- thought leading — best thinking of the best experts
Does your content offer a unique experience?
Define the experiential qualities of your content, the emotions it produces, e.g.:
- empowering — builds confidence
- unafraid of controversy
- clarifying — the bare truth exposed
- aspirational — what you want
- surprising — discover something unexpected
- emotionally inspiring — uplifting
- motivating — seems possible, tempted to try
- calming — made worrying topic less anxious
Does your content show things differently?
- Visual essay — Soak up the scenery (image heavy)
- Confessional — what I learned from my mistakes
- Guided tour by celebrity or expert host
- Behind the scenes at someplace familiar
- On location somewhere unfamiliar — you are there
- Spotlight on — bring attention to something generally in background
- Interview — in their own words
- Myth-busting — The Reality of ________
- Imaginative : What would it be like if…
- Intimacy: True stories of people who _______
Does your content highlight or organize ideas in a special way?
- Lessons learned
- Biographical stories
- Situational anecdotes
- Little known facts
- Explanatory — why things are
- Weird but true stories or fact
- Critical moments: turning point events
- Then and now (continuity and change)
- Below the surface — what you don’t see
- Wise advice — how to live well
Some examples — Do you provide practical advice for people dealing with topic X, or are you a source for breaking news on topic X? Do you have approachable ‘how-to’ articles, or is your content visionary and emotionally inspiring?
Quick Case Study: Netflix “altgenres”
A prime example of a brand using rich and detailed attributes for content is Netflix. Netflix has meticulously analyzed and tagged every movie and TV show imaginable, investing a lot of time to understand how people look for movies. A bot scanning their database found the video service had created 76,897 micro-genres that get super specific, ranging from ‘Scary Cult Movies from the 1980s’ and ‘Japanese Sports Movies’, or ‘Critically-acclaimed Emotional Underdog Movies’ and ‘Evil Kid Horror Movies’.
Netflix specifically sought out ways to differentiate content in their system and find emotional triggers, to surface movies from deep in their database that would easily get lost otherwise, and indulge their users with more binge-worthy recommendations.
“Using large teams of people specially trained to watch movies, Netflix deconstructed Hollywood. They paid people to watch films and tag them with all kinds of metadata. This process is so sophisticated and precise that taggers receive a 36-page training document that teaches them how to rate movies on their sexually suggestive content, goriness, romance levels, and even narrative elements like plot conclusiveness.
They capture dozens of different movie attributes. They even rate the moral status of characters. When these tags are combined with millions of users viewing habits, they become Netflix’s competitive advantage. The company’s main goal as a business is to gain and retain subscribers. And the genres that it displays to people are a key part of that strategy.”
On top of growing viewership and successes of surfacing new and interesting choices to its users, Netflix has an extensive database of America’s taste and preferences in movies and TV shows, and are able to create original shows like House of Cards that are strategically crafted to appeal to what their audience wants.
Set up an emotionally intelligent recommendation solution, then monitor and adjust
Intelligent content is “structurally-rich and semantically categorized that is, therefore, automatically discoverable,” according to Ann Rockley in the Language of Content Strategy. Many brands have a defined voice and/or style guide, and good writers are well-aware of how to incorporate techniques that makes their content distinctive and attractive. By taking this a step further and adding these attributes of the content explicitly as metadata tags, it allows algorithms to generate more emotionally intelligent recommendations.
Instead of relying on big data and user targeting, try building a recommendation engine that looks for patterns and references multiple attributes on content to generate recommendations. For instance, if a person views general interest content with Attribute A tag and Attribute B tag, then show them other content on the same topic with Attribute A tag and Attribute B tag. The closer the match between the qualities of the current content, and recommended content, the more likely the recommendation will be relevant.
Another benefit of including metadata about the qualities of the content is that is allows it to be measurable. Start with a small set of content and experiment, using analytics data to measure what content qualities are in demand, how they perform, and how successful the recommendations are. Adjust the content descriptors and the recommendation matching to see what works and what doesn’t work, and keep an eye out for opportunities to create new content that includes desirable qualities.
To provide emotionally intelligent content recommendations, think about what visitors want from your content, not just what they need. Find general interest content that appeals to a wide range of people, and drill down to determine what distinctions people really care about. Don’t rely on one recommendation model, mix different approaches for best results, and monitor the performance metrics to learn how to improve.
Front-load the work to design a robust recommendation engine and define a tagging strategy that adds distinctive metadata to content — and the good recommendations will flow.