When Discovery Engines Start Acting Human
Written By: Yosi Glick
How content is traditionally recommend
Once video content providers realized that conventional guides weren’t a compelling tool for persuading online viewers to seek out their favorite content, they started adding recommendations alongside the standard title listings. At first, these recommendations were either manually selected ‘Editor’s picks’ or ‘New this week’ titles. This option certainly helped people navigate the sea of choices, but it was an adequate solution at best.
Other recommendation enhancements were later implemented in the guide, such as allowing viewers to tell the guide about their personal tastes (e.g., I like comedies, thrillers or documentaries). Other early enhancements incorporated consumption data to make recommendations seem a bit more personal.
Still ‘Nothing to watch’
These recommendation sets are great for viewers who happen to be in the mood to watch one of the latest blockbusters or whose taste is similar to the editor’s. Nevertheless, many users continue to be frustrated that, despite having an enormous catalog of available content, there is still ‘nothing to watch’. Why? Because the guide itself is too focused on the cataloging needs of the content managers instead of the viewers’ real preferences.
Why traditional models fail
The problem comes down to the decisioning methods used by a traditional catalog model. Content-to-content algorithms called collaborative filtering (‘if you liked X, you may like Y’) have been widely deployed to power recommendation baskets. Collaborative filtering seems to work well in a retail setting, where consumers who bought an iPad may also want to buy a related item, such as an iPad case. But attempts to apply this technique to video consumption produce awkward and confusing recommendations based on statistical correlation, not real similarity. Video consumers want to be shown similar titles, but collaborative filtering is unable to meet this need because it operates in a ‘blind’ mode and lacks meaningful information about the titles it recommends.
When the discovery engine starts acting human
The goal of the video guide should be to give viewers a fun and enjoyable way to quickly find something they want to watch. When people sit down in front of their TV (or any other connected media device) they want to find a program that suits their tastes and mood – right then and there! In order for the guide to enable that, it must make precise recommendations so the viewer can discover the right content.
But creating this type of “discovery-oriented” guide requires rich, accurate information to describe titles in the same way that people would describe TV shows and movies to one another. In other words, the guide must think and communicate like a person and be able to make recommendations based on the viewer’s mood.
Viewers can be in either a passive or active mode. In the passive mode they want to stay safe within their comfort zone and would therefore appreciate recommendations that are similar to their favorite TV shows and movies. A semantic engine can do this well because it understands each piece of content and the user’s tastes so it can then make a perfect match.
When a viewer is in a more active discovery mode, he may want to explore a wider array of available content. The active user will prefer semantic exploration to find the perfect gem. Already familiar with the natural expressions used to describe content, the semantic discovery engine will serve as a good starting point to seed informed discovery across the entire catalog.
Ongoing use of the discovery enhanced guide reveals insights about how the industry makes movies. In such cases browsing itself becomes an enjoyable, enriching experience.
What’s in store – discovery by mood, plot, style and more
Next generation video guides should offer viewers an intuitive way to browse the catalog according to natural language metaphors. For example, if you are in the mood for something stylized and mind bending, the guide should offer this as a way to slice the catalog. You may drill down into your results even further by selecting one or more plot elements, such as ‘uncover the truth’. Users should be able to explore the guide by terms that make sense to them, not the archive manager.
Discovery process should be transparent
An important part of building the viewer’s trust in the guide is making the process completely transparent. When the system recommends a particular piece of content, it should also offer a completely transparent explanation of why it did so. The system should show the viewer what elements the title has that suit her tastes. For example, Boardwalk Empire may be recommended to someone who likes tense, stylized content about gangsters with criminal heroes.
Discovery-powered guide builds consumption confidence and loyalty
Only a guide that is powered with rich, accurate semantic information about each title in the catalog can succeed at recommending similar titles that viewers find useful. A discovery-oriented guide helps people find content they will enjoy and improves their experience, thereby increasing stickiness and consumption. As viewers get comfortable with the service, the discovery-enhanced guide will become a trusted ‘brand of choice’, and they will happily return again and again because they are confident they will find exactly what they are in the mood to watch.
Yosi Glick is the Co-Founder and President of Jinni.
Tweet
Comments