Insider Interview: Gabriel Mandelbaum of Spideo

Web

In practice, content recommendation is difficult to get right (as you’ve probably noticed while scrolling through countless “recommended for you” duds on streaming video providers). There are so many subtleties baked into a good movie or TV show that extend beyond blunt user behavior profiles. Maybe this is why it took an empathically minded French film fan, rather than an engineer, to solve the problem of good recommendation and bring it into the realm of 2.0.

When Gabriel Mandelbaum, founder and CEO of Paris-based Spideo, sought to solve the video recommendation problem, he imagined the hyper-personal experience of visiting the video store in person to speak with the film-obsessed clerk—the one who knows your likes better than you do. Six years later, he counts Canal+, Cenepolis and iflix as clients. But where Spideo really shines is with the content recommendation that transforms Fandango Now’s best-in-class user experience.

Q: You’re obviously a big film fan. Give us your favorite classic director and your favorite modern director.

A: If had to name only one classic director, it would be Jean Renoir, particularly when he directed The Rules of the Game. It’s poetic, funny and sarcastic, and honestly faces the world as it is. For modern filmmakers, I enjoy the intensity of Scorsese’s DiCaprio period. I was in my twenties when I first saw Gangs of New York, The Aviator, Shutter Island and The Wolf of Wall Street, and each one of these movies resonated immensely with me.

Q:   What were the milestones to getting Spideo’s recommendation function to where it is now?

A: At the very beginning of Spideo, we thought that the digital video world was missing something. We were missing the video clerk who knew me, who had seen all the movies, who could talk about it and guide us through the hundreds of DVDs available.

We think that the best recommendation experience is the one between two human beings. It’s amazing how creative people become when they describe, explain and convince someone to watch a particular film or TV show. We identified this type of conversation as the ultimate recommendation experience: A guide through thousands of movies available, asking simple and limited information from the user. So the ground zero of the algorithmic process has been to exhaustively analyze how people talk about movies and TV shows.

Mandelbaum3
Mixing moods with Spideo on Apple TV.

We built our initial ontology to make sure our algorithms would understand and describe content with the very words people naturally use in their daily interactions. Based on this first version of our taxonomy, we have built algorithms that would simply reflect the way people discover or decide how to watch a movie: according to their mood, based on a film they have previously watched, based on their preferences and their viewing history over a longer period of time. Since then, it has been an ongoing process to improve the semantics and the algorithms and the way they interact together to provide the most relevant recommendations to users.

Q: What is your current ratio of algorithmic to editorial curation, and how do you see it evolving?

A: We never see the equation this way. Instead of a ratio, we use a combination of the machine and human elements. The algorithm would not be able to do a good job without the curation and the human eye. The algorithm goes with curation, but we call it metadata creation, or semantic enhancement. We spend a lot of time refining the process, but it is not one against the other.

So human experts work hand in hand to create automatic indexing with machines. The automation exists to improve the automatic indexing, but not to remove the human eye. It allows us to manage a greater amount of content every week and every day. We really think this is what makes Spideo what it is. We make these two elements work together, not compete with each other.

Q: As artificial intelligence becomes more sophisticated, how do you envision its role in Spideo over the next three to five years?

A:  What we’re seeing is that machine learning is actually helping us to go forward and move to a new kind of business—to not only movies and TV, but to every type of video content. For TV and movies, you have a lot of metadata and you can find very good descriptions. But when you’re facing unstructured video with no metadata at all, when you’re trying to go very deep into the content, this is where the intense learning is going to help us a lot. What we are trying, and learning, is to, for example, index gaming videos like the kind you can find on Twitch.

Spideo screen shot Apple TV.
Would you rather?

Q: What is the place of the set-top box in this context?

A: We have done a lot of work to make our API simple for a set-top box. I think the set-top box should evolve to become a visual proof of quality across content, recommendation and ultra-high definition delivery. Then, if people pay for greater service and see it physically in their living room, it works. But if they’re paying for the service and the UI is crap and they have no recommendation, it will not survive in the long term.

Q: We’ve seen a lot of big acquisitions lately. I am always curious to get an insider’s take on this.

A: What I can tell you is that I think everyone is looking for a good product, and for the right audience to present it to. I am most excited when I see a large company buying a tech company and trying to innovate. So when you see AT&T buying QuickPlay or Comcast acquiring Icontrol, there are a lot of little pieces coming together, and these companies are putting a lot of money into making it work. They will have a very compelling product at the end, and they all want to compete directly with Apple and Google.