VideoSurf’s New Release Jun 10, 20097:59 PDT

VideoSurf’s latest release is out and we’ve got some exciting new developments to share with you.  In addition to the updated look and feel, we made some big improvements to key areas including search relevancy (spam detection, video deduplication), people spotlighting, and our homepage.  This new release is all about helping you find what you’re looking for, faster.  Whether that’s a search result, a celebrity, or a cool video to watch, we’re working harder to make sure you spend less time searching and more time being entertained.

Improved Search Relevancy

Advancements in the Detection and Treatment of Spam and Video Duplication
Search is what we do here.  It’s what we have a team of world renowned mathematicians constantly iterating on and tirelessly working to improve.  Our goal is always the same: better, more relevant results for your searches and a faster way to connect you to the videos you want to watch.  The improvements we’re introducing with this release in the areas of spam and video deduplication go a long way towards making game changing steps forward in the area of search relevancy for video and we’re excited to share with you what we’ve built!

Attacking Spam
Based on our research we estimate that spam videos comprise more than 20% of the videos on the web! It’s no wonder that users spend most of their time on other sites skipping back and forth between the search result page and the video page just trying to find what they’re looking for.

VideoSurf’s Visual Summaries have gone a long way towards eliminating this hassle, but our latest release takes our fight against spam and our quest for superior relevancy to the next level. VideoSurf has now improved the way we use our computer vision technology to automatically detect these videos and algorithmically devalue them in our rankings before they even make it onto the search result page. Because our technology can process the video (instead of just the text associated with it), we’ve taught our computers to visually sort out the good from the bad so that our ranking algorithms can use that information to deliver users the most relevant results.

Here’s an example of how it works. Often times spam videos will strategically place a misleading image in their video (ex. a Victoria’s Secret model) so that sites will pick it up and use it as the main thumbnail displayed on the search result page. They do this to encourage click through and to fool users into thinking their video is about something it’s not.

Here is an example of a video with a misleading thumbnail…

Because other video sites are unable to see the video content, they can’t know that their users are being misled. VideoSurf however, is able to visually identify the problem and boost the video down in our search rankings automatically. We even take it a step further. If we need to show this video in our search results, VideoSurf is able to automatically correct the inaccurate thumbnail so that the leading thumbnail we show is more representative of what the video is actually about (all this in addition to showing you the most important scenes in our Visual Summary).

Here’s how this video shows up on YouTube’s search result page:

And here’s how it’s displayed on VideoSurf’s search result page:

Apparently, when you look like Professor Albert, a fair amount of click-bait is required to get your video watched; fortunately, the VideoSurf technology doesn’t let him get away with it.

Our computer vision advances in spam detection help to ensure that our users see the most relevant videos (with the most representative thumbnails) front and center for their searches.

Getting Rid of Duplicate Videos
Our testing has shown that for the top video search queries, over 85% had duplicated videos returned on the first page of results. We know users are frustrated by having to scroll past the same video over and over again to find what they’re looking for.  So, with our latest release, VideoSurf’s Computer Vision technology was put to work to better recognize duplicate videos and group them together to improve search relevancy and save time.

Other video site’s deduplication technology relies on matching up titles and start/stop times. It’s not their fault; that’s all they have to work with. However, our ability to see inside videos lets us take this a step further. VideoSurf is able to identify each video’s unique visual DNA and match it against others in the index.  Even for videos that don’t have the same start and stop time, title, or metadata, if we can see that the visual makeup is the same, we can identify it as a duplicate.  On our search result page, you’ll now see duplicates collapsed under the main video, so you can scroll right by them (or expand the section to see how many we’ve found).

Take a look at the example below, where the fourth result for a search on “Thom Yorke” has 10 duplicate videos associated with it.  Even though the titles, run times and descriptions are all different, our ability to recognize the similarities in the visual content is what drives the deduplication.  Especially for top search queries like celebrity names, song titles, TV shows, etc., video duplicates have a huge impact on search relevancy, so we’re very excited about our advances in this area.  Try the same query on other video search engines and see what you get (we warn you though, it won’t be pretty).

The VideoSurf Spotlight Effect

One of the new features on our updated homepage (which you’ll hear more about below) is the “People in the Spotlight” section, which links to videos of the most searched for people on the web at the moment, as found by our computer vision technology.  Oftentimes though, the person that we’re looking to highlight is pictured standing in a shot with a group of friends, or obscured by a busy background, or hiding from the paparazzi.  This is where our “Spotlight Effect” comes in; to shine a light on the person we want to call out while deemphasizing the extraneous parts of the photo.

Take a look at these two images to see the before and after. Not only can our computer vision technology automatically identify which Jonas brother is which (no small feat!), but it also has made it easier for you to find the one you’re looking for by cropping the photo to best feature him, highlighting his face and fading the rest of the Jonai into the background.

We know that a picture is worth a thousand words so we use this technique across the site anywhere that our visual capabilities identify and let you search by a character.

In addition to the homepage you can also see it in effect on our search result pages:

As well as on our Video View pages:

Compare this VideoSurf feature to Bing’s use of “Related People” on their video pages. Because the Bing engine can’t see what it’s showing you, 8 times out of 10 the person being “featured” didn’t even appear in the thumbnail photo!

New and Improved Homepage

We started out with just a search bar, but we’ve added some new video discovery options as well to help out those of you that are stumped on what to search for.

•    Hot Searches features search terms that are among the freshest and most searched for across the web.  Culled from search information gleaned from text based sites (like Google and Twitter) and video sites alike (including our own search activity trends), if you want to see what’s going on in the world, simply click on one of the Hot Searches we’ve selected to find the most recent videos to watch.

•    People in the Spotlight is a collection of the most searched for people on the web at the moment.    This section is constantly updated and features the celebrities, politicians and average Joes that people are currently talking about, which our computer vision engine has recognized (and spotlighted!).

•    WatchBoxes feature some of the most interesting, popular, and watch-worthy videos for each of our top categories. If you’re interested in Movies, TV, Music, Viral Videos, Sports or Celebrities (and we know you are), WatchBoxes are a great way to skip right to the good stuff and start watching.

Stay Tuned…

Video Surf has grown to include over 50 billion visual moments in our index and we’re adding more every day. The list of top video providers we index now includes USA network, SciFi.com, the fashion and modeling site Modelinia, and more, adding depth to the videos provided by mainstays like YouTube, Hulu, Comedy Central, and MTV. All told, it would take you roughly 400 years to watch all of the video in our index… not counting the videos that have been added in the time it took you to read this!

We’re continuing to improve VideoSurf to help you sort through everything that’s out there and find the clips you really want to watch, so keep checking this space to see what’s next and in the meantime, start VideoSurfing!