How Storyblocks uses machine learning to offer Recommended Music for your videosProduct Updates
October 7, 2022
How Storyblocks uses machine learning to offer Recommended Music for your videos
As Storyblocks continues to invest in exclusive demand-driven stock media libraries, our data scientists are leveraging machine learning to surface recommendations across content types. These recommendations help you find what you need to test different creative, and produce more videos faster than ever before. In this interview, we chat with data scientist Zach Lamberty about his role and the insights that fuel these innovations, as well as one of our newest features — Recommended Music.
What makes Storyblocks such an appealing place to work as a data scientist?
“If you talk about what’s happening in machine learning and data science, people are most interested in video, audio, images, and computer vision. Text and NLP (natural language processing) are what they’re interested in. To tie that back to Storyblocks, we want to help creators do what they need to do better than they could before. At Storyblocks, we have all of those data sources, along with a great way to surface recommendations. So, our users get to see our work straight away.
“Because our business model allows for unlimited downloads, we have a really strong signal that would be hard to get otherwise. At Storyblocks, so many things about the problem space are set up really nicely for a data scientist. And of course I love my team and the people I work with. So a lot of great reasons.
“When I was first working as a consultant in the data science space, I worked with a lot of government agencies and private companies. Every time I went into a new project, I was just hoping they had good data, so we could do interesting things for them. So, when I came across an opportunity at Storyblocks, what appealed to me most was that we have a really amazing set of data — that’s really hard to find.”
What are the most interesting insights you’ve found about Storyblocks customers, and how has this driven product innovation?
“One of the early learnings for us was that the way that people search for video and the way they search for audio are just intrinsically very different, and that it’s often a huge struggle to search for audio. Unless you’re an audio expert, it can also be a huge struggle to talk about audio. You might know what you want, but it’s hard to put it into words. And that means finding the right audio content can take a really long time. So, that insight has been fundamental for us and how we approach search, recommendations, and data science.
“We always want to strike the right balance between where people want assistance in their search, and where people want creative freedom and ownership of that experience. That feedback from our users and creative experts is incredibly valuable, because they’re our stakeholders. We wanted to learn how they build projects, and what’s easy versus what’s hard. The consistent feedback that we hear through all of those outlets is that people like to select their video content first. But we also hear that knowing where to start with music is a pain point.
“All that led us to launch our “Recommended Music” functionality alongside the videos users see in their search results. Recommended Music helps users quickly find the best tracks for their creative project. All they have to do now is hover over a footage clip in their search results. Then, click on the “Recommended Music” icon. Once you do that, you’ll find three songs that pair well with the footage. Ultimately, we want you to choose the audio that works for you and your project. To help you get started, we’ve narrowed the universe from like 30,000 options to just a few great choices.”
What data are you using to make music recommendations for Storyblocks users?
“So, in any robust machine learning or data science project, you’re looking to leverage information that real users give to you. They’re your experts. You’re trying to learn how to reproduce the actions or opinions of people as faithfully as you can, at scale. In the process, hopefully you learn something that you couldn’t learn by just asking people a question. Searching for music and video is a core part of our users’ experience. So, we’re fortunate to have a number of ways to get at this valuable data at Storyblocks.
“Our most affirmative way to see how real people find a piece of audio that fits with a piece of video they want to use, is by looking at the finished projects they create using Maker — our online video editor which is fully integrated with the Storyblocks stock media libraries. So, Maker gives us these storyboards from users who have chosen specific music and video clips to produce their projects.
“Additionally, we also provide Multimedia Folders on our platform. People use these to start thinking about their projects in a multi-asset context. So, they’re choosing a variety of videos and audio that could work well together, and they’re grouping them together there.
“All of these data sources together are a huge asset for us, and one that’s almost impossible to replicate. We can recommend music tracks with a high degree of certainty that they’ll pair really well with a certain video clip. That’s because Recommended Music is based on the aggregated decisions that we see creators making across our entire platform. And we do this all while prioritizing our users’ data security. We never store sensitive user information in our data applications or machine learning models.
“We’re excited to use this information to help all of Storyblocks’ customers find relevant music. It will help them work faster, and hopefully delight them throughout their entire creative process. This is just the start of many more recommendations we’re working on so check back in soon.”
Try Recommended Music for yourself
Want to see Recommended Music in action? Try searching the Storyblocks video library today!