Modern Ways to Maximize Metadata for Better Content Discovery and Recommendations
It’s not going to come as a surprise to anyone that high-quality metadata is essential if you want your content recommendations to be successful in boosting streaming engagement. Equally, I’m not betraying any confidence when I say that every streaming service provider I speak to finds metadata to be the biggest challenge in their personalization efforts.
Manually adjusting and enhancing your existing metadata to make it more useful for recommendations is time-consuming and expensive, even if your library is relatively small. Buying additional metadata from third parties is sometimes suggested as a solution, but this is also costly, plus metadata is an excellent example of an occasion where more doesn’t necessarily mean better.
Making better use of a smaller range of keywords
If every content item in your library is associated with 40 different genres or sub-genres, then ou’re not going to get particularly well-targeted recommendations! Conversely, having a wide range of keywords that are rarely used is also unhelpful. As I mentioned in a previous article for Streaming Media, I once had a customer boast that their content management system contained more than 50,000 unique keywords. But if a keyword is used only once, then that’s not going to help you match it with any other piece of content.
Consistency is also very important. When digging into the metadata of one customer, we found that they had the first three Die Hard movies in their SVOD library but their metadata really wasn’t helping to tie them together. While Die Hard had been categorized as a “cop movie,” Die Hard 2 was a “detective movie,” and Die Hard 3 was a “police movie.” To you and me, it’s clear that these are all the same thing, but algorithms are notoriously hard on this kind of stuff. They just don’t see the correlation between these terms, and if they’re trying to find the ideal match for someone who has just watched Die Hard, you can see that this use of keywords is not at all helpful.
One thing that might help in this scenario happens to be another common failing that we see across many streaming companies. They lack the concept of “franchise” in their metadata. While the hierarchy of series and season is common, it’s not always easy to identify other shows or movies in the same family. For example, if someone’s just binged every season of CSI: Miami, you’ll probably want to encourage them to start on something else from the CSI family (CSI: Vegas, CSI: New York, etc.), if it exists in your library, because they’re already primed to enjoy it. But without links between different series or movies in an overall franchise, that’s much less likely to happen. They may just get a recommendation to watch another crime drama, which is useful, but not as targeted.
Rise of the machines?
So, what’s the solution to recommendations that are more intelligent than simply “You watched Top Gun, here’s another movie with Tom Cruise in it?” We’ve seen real success with a mixture of machine learning to automate the process of updating hundreds of different content records in a cost-effective and timely manner; and a focused amount of manual intervention.
Let’s be clear, I’m not proposing that you let bots write your content descriptions. However, with the right models, it is possible to significantly enhance your use of keywords, which are what most content recommendations rely on.
It’s a multi-step process. First, you’ll need to analyze your existing metadata to see what kind of keywords you’ve already got. Then you need to define a manageable framework of preferred keywords—including genres, sub-genres, franchise and even mood-based categories. Manageable is the essential word here. Too many or too few keywords will make it impossible to keep track and will diminish the quality and number of matches you can make.
A metadata framework for the future
There is no one “correct” framework or schema that can be applied across all streaming services. It needs to be crafted to suit the type of content you’ve got (or plan to get in the future), your audience, and the tone of your service. Once it’s been defined, you’ll then need your data scientists to build an advanced metadata-prediction model to “fill in the blanks” in your metadata.
Now comes the really tricky bit. This new framework of keywords you’ve defined needs to be used consistently across your entire library. Not just now, but on an ongoing basis. Our experience with streaming services shows it’s important to have policies, processes, and people in place to ensure this continues to happen. All new content will need to be tagged with keywords that follow your agreed framework.
Of course, you’ll also need to be ready to adapt the framework to accommodate changes in your business over time—for example if you acquire your first sports rights or branch out into kids programming. Like everything with content personalization and engagement-building, metadata enhancement is not a “once and you’re done” process. But with judicious use of machine learning you can make it an easier process.
Metadata matching and new methods of content discovery
One of the great things about working with different streaming services is that the 24i data team gets exposed to all sorts of interesting ideas about how to improve the content discovery user experience. One of the coolest we’ve been involved with recently was enabled by our machine learning metadata enhancement.
One of our customers was keen to help its consumers find more of the content in its enormous VOD library. We helped them to define and apply tags that match each of their videos to a selection of keywords based on a frame of mind—like “cheerful,” “captivating,” “moving,” or “thirst for knowledge”—as well as more traditional genres and subgenres like “crime,” “action,” or “romance.”
A special section of their interface now asks consumers to select a category that best describes what they’re “in the mood for” and to specify how long they want to watch for (<10mins, <30mins, <60mins). Crucially, our APIs also enable a “daily shuffle” feature to ensure the order of the content selection returned is kept fresh, even if you’re in the same mood for several days in a row.
This is just one example of innovative ways that service providers can improve the content discovery experience if they get their metadata right. It’s also a great way to showcase content to new users who haven’t yet built up a viewing history that makes personalized recommendations relevant.
Another strong side benefit of enhancing your metadata is search, which is often a little neglected by OTT providers but can account for 20% of all plays. When you enhance your metadata, all these new keywords and franchises you have added are then available to be searched. So if you search for ‘dark comedy,’’ you get Daniel Sloss without knowing he is in that genre. Searching for a franchise is also very common including production brands (e.g., “Disney”) and sports competitions and teams (e.g., “Champions League”).
Moreover, many TV platforms are also starting to see strong growth in voice search. When using voice search, people are much more likely to use it in discovery mode (e.g., “show me something uplifting”) rather than a specific title they want. When you have enhanced your metadata, not only are recommendations an order of magnitude better, but your search wows your users. This all combines to give users an amazing experience and to find something great to watch easily.
You’ll find more on how our customers are benefiting from enhanced metadata in the 24i e-guide Five engagement-boosting strategies every streaming service should adopt right now.
[Editor's note: This is a contributed article from 24i. Streaming Media accepts vendor bylines based solely on their value to our readers.]
Personalized content recommendations are a hot topic in streaming. Extending them to your emails and other marketing messages is a logical way to increase repeat visits to your apps and counteract churn.
Many streaming services have got binge-watching nailed for episodic content, keeping viewers hooked on their hottest series with auto-play of the next installment. But are you wasting a golden opportunity to recommend new content at the end of a series or when a viewer finishes watching non-episodic content?
If you're looking to add or upgrade your data-driven recommendations, should you go for Recommended For You, Because You Watched, or Trending to bring the greatest increase in engagement? 24i's SVP of Data Products, Damien Read, argues the choice of section names and the data models behind them should be tuned to your user base and your specific business goals.
Going viral in the streaming space is a taller order than when audiences first coined the term "binge-watching" as streaming platforms started dropping full seasons of content all at once. But what if, even in an increasingly crowded space, content owners and buyers could use data to determine whether content will drive viewership—or that it might even be binge-worthy?
Amid fierce competition, streaming services are looking to new personalization strategies to increase content engagement, tackle churn, increase advertising revenue, and encourage users to return to their apps. What new and emerging data-driven approaches should streaming services use to boost their content discovery user experience?
Companies and Suppliers Mentioned