Making Learning Searchable with Mobento

Originally published on the Clarify.io blog. View archived copy.

About 18 months ago the video learning company Mobento came to us to ask how to make their content searchable. This included over 10,000 hours of educational videos. At the time, we weren’t sure.

We had one major advantage: most educational videos are fairly high quality. We were used to processing telephone calls which are poor quality by comparison. The technology used to extract words from these two types of audio is different even though the underlying algorithms are fundamentally the same.

Working together we were able to apply our technology to their content. The Mobento you see today is the result of this work, and thanks to that Clarify now handles both types of audio very well.

 

Before going any further I should be clear about their goal. Mobento’s videos were already searchable the way most video content is today: using metadata, such as speaker, topics, descriptions, etc., which are manually entered. But they also wanted to index these videos by their actual  content, i.e., the words spoken.

This was new.

Why did Mobento feel the need to index their library in this way when no one else did?

Their answer was simple: YouTube. The students who use their site love to consume video, but they like to consume it in 3 to 5 minute increments. If they’re doing research, or reviewing for an exam, the last thing they want to do is (re)watch a 90 minute lecture. They want to skip to the bit of the lecture they care about.

Mobento knew their users and realized they only had two options: manually tag each section of each video or find a way to do it automatically.

The manual option was far too expensive and time consuming. They planned to create a library of 30,000 curated video within 24 months. They couldn’t afford to have people tag all that content and keep to their schedule. Using an early version of our API, they could.

After a few months of testing, we were all satisfied with the quality of the results, so it was then just a matter of connecting their CMS to our indexing system, and building a player that could display the result of their users’ queries.

Modifying their CMS workflow turned out to be trivial. Building a player capable of displaying inline results and allowing the user to cycle through them was more difficult, but they managed, and the result is stunning:

In the example above, the user searched for “solar” and “electricity”. Those terms appeared in the audio and in the description. By clicking on the result labels, a user can cycle through each of the results.

Before Mobento introduced this feature, most of their users found videos by clicking through a topic hierarchy. Full search (audio content & associated data) was meant to enhance this navigation, but it went beyond that. Full search made hierarchical navigation obsolete.

Predictably, Mobento’s users were enthusiastic about the advanced search feature, but what really surprised everyone was that these users stopped using topic hierarchy navigation altogether. Mobento mistakenly assumed that their users would use both. Once it became clear that they wouldn’t, Mobento eliminated the directory-style means of discovery from the site.

For those old enough to remember Google’s entry into the search market, this isn’t surprising. Before Google, Yahoo! was the most popular web page discovery site on the Net. They hired humans to categorize web pages and organize them in a hierarchy. Google Search made that work obsolete overnight.

Mobento doesn’t want to index all of the video content on the Net, but they’ve certainly shown the most effective way to make that content discoverable, and we’re proud to have been able to help with that work.

If you have any questions or would like to get in touch, feel free to comment or tweet us: @OP3Nvoice.