YouTube's Moodwall uses feelings

New YouTube browsing tool sorts videos into the emotions they invoke The Moodwall has 20 emotion categories, including sad, brilliant, funny and epic Project is latest attempt by YouTube engineers to solve common discovery problems (CNN) -- YouTube is getting in touch with its emotions. The video service rolled out Moodwall on Friday, a new way to discover and watch videos based on how they make viewers feel. The experimental page divides videos into categories (called Vibes) such as amazing, cchy, scary, gross or intense. If you're feeling melancholy, you can turn up The Smiths and browse through the sad selections. If you want happier tears, click Powerful; for a good giggle try Funny. "Videos aren't just about objects and things, they're about the emotions they invoke," said software engineer Sanketh Shetty who worked on the project. To find out what emotion to assign a video, Shetty and his team turned to an unexpected place for help: YouTube commenters. YouTube gathered data from the comment threads on videos to identify common words and phrases, detecting what response the video had inspired in viewers. The project is the Google-owned company's latest attempt to improve how viewers discover videos and addresses a specific problem. Engineers noticed that when people searched for content to watch, they were putting in extremely general search terms, such as "Funny video," which tends to return videos that have the word funny in the title or description. "The broad query problem is pretty well known at YouTube," said Shetty, who has worked on discovery issues in the past. In December, the company announced YouTube Slam, which pitted popular videos against each other, allowing viewers to vote on who was a better dancer or which kitten cuter. Moodwall takes this idea of crowdsourced groupings one step further. Around 40,000 videos have been put into 20 emotion categories. Each time you visit the page you're presented with a random grid of 15 moods and large thumbnail images, but no video titles. When you click through to a mood, you see another grid of thumbnails with secondary emotions that might be associated with the videos (for example, sad videos can be sad and amazing or sad and beautiful). The video for the original thumbnail image you clicked is displayed larger and shows the video name. Once you choose a video, a playlist bar will appear at the bottom of the screen. Videos for that feeling (or combination of feelings) will play continuously, and viewers can jump around to other videos in the queue. The page was tested with a limited number of YouTube users over the past week and is available to all U.S. users starting Friday at youtube.com/moodwall_testtube. "We want to give this free flow, associative browsing experience," said engineer Madison Le, who worked on the project's clean interface. Moodwall was created by a team of four engineers in two months. The team, which calls itself Team Jeannie because they "grant our users wishes for what they want," didn't leave the results up to commenters entirely. They looked at other streams of data, including popularity, and then hand-curated the moods. As anyone who has read YouTube comments can attest, curation is key for filtering out the noise and making sure there is not a "First!" emotion. But the unique voice of YouTube users is still present in Moodwall. For example, there is an Epic mood category. "You have to be open to the possibility that YouTubers will have different ways of describing things," explained Le. The team will continue working on the Moodwall project after this release, listening to user feedback and rolling out new features. This is not the first time YouTube has sifted through comments to gauge viewers' reactions. In a research project earlier this year, Shetty created an algorithm to determine how funny videos were based on keywords in the comments. He looked at the volume and types of LOLs (all caps, length of looools and amount of exclamation points were all taken into account), as well as other funny words including "Haha," "ROFL" and emoticons. if(typeof CNN.expElements==='object'){CNN.expElements.init();}