Search, Recommendation, and Curation: Attract and Keep Viewers

As any publisher knows, “Post it and they will come” isn’t an effective strategy for getting your video seen. The latest strategies in search, recommendation, and discovery can help your content stand out among the rest.

As video distribution and consumption continue to grow exponentially, viewers need efficient ways to find what they’re looking for. In an extreme example, a recent Google search for “cat videos” returned almost 400 million hits. How can publishers make sure that just the right cat videos get in front of the people who want to see them?

Viewers only have so long an attention span, and you don’t want them wasting it on the wrong content. The problem with video is that it’s essentially a blob, a fairly unsearchable (or what engineers like to call unstructured) piece of data. Viewers need help finding what they want, whether it’s a particular video in a huge library of content or a particular segment within a single video.

Search, recommendation, and curation are the three keys to getting viewers to the content they’re looking for and helping them find content they might not even know exists. Search and recommendation are primarily automated, algorithmic events. For the purposes of this article, search refers to finding something within a particular video, while recommendation offers viewers guidance to finding the videos. Curation adds a human touch to a technological workflow that presents viewers with the best content around a particular subject or theme.

This article looks at how six companies are working to solve the search, recommendation, and curation challenges.

Search

Technology companies Ramp and Panopto have search tools within their platforms to enable you to find content at a specific point within a piece of video. They do this by developing very rich metadata based on time-coded transcripts and a complete set of thumbnails. This detail allows a search engine to find the exact location within a video for the term a viewer is looking for. When video is richly tagged, transcribed, and titled, it has a much better chance of being surfaced in various content systems and by search engines—and therefore seen.

American Public Television (APT) wanted to make it easier for viewers to find programming online from within a large library of content available from its CreateTV series. “The primary business we’re supporting is broadcast television. We’ve created a companion website (createtv.com) to provide channels a rich website they can also use for engagement and retention of viewers,” says Gerry Field, vice president of technology at American Public Television. “We wanted some way to search through programming and help people navigate it.”

Videos on CreateTV.com are accompanied by full transcripts that allow viewers to search within a video, as well as selected terms in a blue bar that have particular significance to the program. For travel videos, the webpage includes a Google map triggered by the content of the program.

Field says the CreateTV website uses a version of the JW Player that’s combined with Ramp’s transcription and search functionality. “What you’re seeing underneath [the video] is a full transcript based on captions that we choose to display under the program,” Field says. “You can also land on any part of the transcript and it will go to that part of the program. In the blue bar underneath the player are terms we’ve identified that might have significance to the program.”

A new feature APT has recently started to use within its Ramp system is a rules-based trigger that can display related content or advertisements external to the player on the webpage.

“We have a very simple thing here that we wanted to do for travel and that is give some context to the video. The Google map,” which Gerry identified on the right of the player on the webpage, “is triggered by the content of the program.”

“Any time you have a lot something you have a search problem,” says Tom Wilde, CEO of Ramp. “When you think about search in general, most search systems have been built around an assumption of text. Google, Facebook, and Twitter all are based on text as the primary raw material.”

But video is a binary file object that’s opaque to traditional or enterprise search technologies.

“So our strategy is to wrap each video with as much metadata as you can, and ultimately you have what appears to the search engine as a document,” Wilde says. “In many ways we say video is the new document because it’s taking the place of so many use cases the documents occupied.

“So what it means is the right piece of content in front of the right user, at the right time,” Wilde continues. “If you put all this effort into producing the content and people can’t find it, it’s of no value to anybody. Making sure that part of your video search is not just finding the video, but finding the point in the video that’s relevant and providing this jump to functionality is something else that’s different about video, but it’s also something that users expect.”

In addition to supplying as much text as possible with your video, Wilde recommends using thumbnails—which he says dramatically increase click-through rates—and site maps. He recommends a complete video site map, one that’s submitted to the Google Webmaster site map too.

Panopto compiles audio transcriptions, onscreen text, and metadata to provide a rich collection of searchable terms and strings for the videos in its system.

“Google has a separate site map for videos, and you definitely want to do that because that’s how your videos show up inside Google as a strategy for web discoverability,” Wilde says.

And even though much of what companies such as Ramp offer is achieved via automation, technology alone can’t provide all the solutions. “I’m really aware that providing captions when you don’t [already] have them is not a small thing,” Field says. “It needs to have eyeballs on it; it can’t just be machine-generated. With travel programs, we have programs from all over the world. Voice recognition typically doesn’t do that well.”

Qualcomm uses Panopto’s tools to set up search strings to find content within its video library to create a range of corporate training, sales, and partner engagement services. The Qualcomm use case is one in which a viewer goes into an archive to find something very specific and get it out quickly.

“The ability to search for a specific talking point, in the slides or notes or anyplace a speaker has said something that matches the search string, that’s pretty cool,” says Ken Davis, learning technology specialist at Qualcomm. “Panopto provides one-click recording, uploading, and distribution right out of the box. That was really appealing to us. What we’re trying to do is recording of instructor-led events. It’s much more of a learning engagement tool than just a video streaming platform, and that was the appealing thing to us in the learning center.”

Davis says his group does quite a bit of curation, compiling relevant videos for particular stakeholders. “We send those links out as learning items so somebody can go to that link and see a whole list of different videos that contain the search string,” he says. “Each viewer can create notes which then are synchronized within the video and become part of the content and can be shared with other people.”

“We realized to solve a problem of video search we had to help people create content that was intrinsically searchable,” says Panopto CEO Eric Burns. “Video is very challenging to search because it’s just a sequence of images and also has very large file sizes. So unless it’s decorated with metadata, unless there’s some sort of curation, in a lot of cases, it’s dark content.”

The challenge is finding something within 150,000 hours of video content. Panopto built its product on the understanding that many presentations use slides to structure content. Its product does full speech recognition on the audio track, runs optical character recognition to extract any onscreen text, and indexes manual metadata, such as descriptions, titles, creator names, and creator bios.

“One of the key pieces of Panopto search is that we just don’t find you the video,” Burns says. “We actually find you the instant in the video where your search term occurs.”

So if you have a 3-hour recording of a sales conference or a town hall meeting, Panopto’s search will allow you to find a specific reference within the video.