YouTube topics
There are forty-eight hours of video uploaded to YouTube every minute, or eight years of content uploaded every day. In Hollywood terms, this is the equivalent of 240,000 full-length films uploaded every week. The question this raises though, is how do people sort through this massive amount of content to find what they want to watch?
On approach is to give people lightweight tools to help them express what they're interested in. We wanted to encourage and facilitate exploration of videos through a system of concept building blocks. We knew there were a considerable amount of browsy queries on the site and that the need was there. The core concept came from an engineer who I was working closely with. We spent a bunch of time together hashing through the details of the idea to create a system and we created a stop-motion video of the concept.
We thought about topics as a language that could help people express what they wanted on the home page, on the search results page and on the watch page. On home, they could be helpful in prompting people with interesting or trending topics (e.g., london, riots, rebecca black), and they could be used to create finer tuned subscriptions (e.g., I want to see new content from the BBC and the NYTimes on design and business). On search, they could be used to express compound topics (e.g., trending music parodies), or even help for navigational queries when a user can't recall the specific title (e.g., mom, sneezing, baby). On watch, they could help a person understand the pivots they could take from a specific video e.g., from Pompalmoose's cover of Single Ladies, some pivots could conceivably be: popular, indie music, covers, or beyonce.
We shopped around the concept of topics and convinced enough folks that it was an interesting idea to assemble a strong team of engineers. At the infrastructure level, the engineering team began to link topical relationships from a wide variety of sources and began to associate topics with specific videos. While they got to work on the infrastructure, I started to create a visual system and work through the interactions in more detail.
The first visual explorations were inspired from our building block idea and while they looked great in wireframe form, completely detracted from the core task of searching and looking for a video to watch. I did a ton of variations on this, but these are a few of those approaches. Visual design iterations will be added soon.
The trickiest set of interaction details I worked through were around the introduction of topics into the search box. At the conceptual level, there was the question of how we should deal with user-inputted text that could also be interpreted as a topic? How should we deal with disambiguation? Should topics add to or replace an existing query?
We originally assumed that topics should be additive. A person could search for "amazon," then add (deforestation) or (kindle) and we thought this would help people form better queries. Obviously the down side is that as you add more concepts, you narrow your results further. When we ran the protoype through a first round of user studies, most people quickly got them into a state where they had no results, had over-filled the search box, and didn't know how to clear the search box. If we had chosen to support replace only, we would have reduced a big part of the power of the concept, and there were also a class of topics that could only be additive e.g., (high-definition) and (closed-captioning), because these topics were meaningless on their own.
We decided to proceed with default-replace model that accounted for a subset that were additive-only and also supported a harder-to-find power user feature that would allow people to create compound queries. Everyone on the team contributed interaction ideas as we progressed. One really nice subtle touch that an engineer suggested was the use of light gray text in the title bar to foreshadow what was going to happen to a user's query.
We launched a public opt-in for topics, ran a series of 1% experiments and conducted research to gather additional interaction and qualitative feedback. We were really excited about the power behind topics, but people just really weren't interested in creating the perfect result set. We did do a full launch of topics as exploration and refinements on the search page (currently live). The team then started looking at these videos aggregated around topics and started to produce some really high quality pages for topics. Given that the quality is actually pretty good for a huge set of topic pages, we're now looking at how to integrate these into the existing site architecture, where the best touch points are, and what the experience should be for interacting with topic pages. To see topic pages, go to YouTube's labs area (TestTube) and opt-in to Topics.
I was the designer on topics. I worked most closely with an engineer manager, who had the original concept behind using building blocks for queries. As the project kicked off, I worked closely with the engineering team (back-end, front-end and web development), the PM and an amazing project manager, who kept the 1000 moving pieces and all of us in check. I bugged a visual designer for feedback from time to time and she created the current look and feel for topic pages.