Content data at scale : the underrated Eldorado for video platforms

Paul Chaumont
May 27 · 6 min read

There was a time when assessing video content data was manageable by humans. With a constant but reasonable flow of daily videos uploaded or published, any video platform could assign the “tagging” part of content to a dedicated team or by users themselves and ensure having minimum information to help them sort their inventories and sell them to the most offering.

But times have drastically changed:

· Video has become the N°1 medium all over the internet. As it is said that in 2020, video will represent 80% of internet traffic, figures quite demonstrate the hugeness of video market. Even though impossible to measure for the whole internet, YouTube figures only are mind blowing (I.E: 300 hours of content uploaded every minute). Video content has clearly gotten out of human control.

· Users attention is volatile and content quality is decreasing. Would you remember the last video you watched ? Probably not, as it might be the 20th of the day, mostly user generated content, sometimes with low quality and for most of it, less than 90 seconds. Video production has stepped into a new era: it is not a professional privilege anymore. This also mean that targeting users with content has become a tricky game, mostly because control over content itself has disappeared. A brand or a celebrity might as well appear in a homemade video than a producer’s.

· Manual tagging has proven itself erratic. We already discussed this matter in YouTube’s case: manual tagging is facing a great deal of either inaccurate information or ineffective process. Most cases, videos are flagged with inappropriate and partial tagging, preventing content providers to make a commercial use of it.

· User data is on the edge. Due to GDPR regulation and ever-growing user reluctance to share their private data, targeting has become hell. Multiple profiles and email accounts, false information, ad blockers, unfriendly cookie settings are as many reasons to explain the difficulty it has become to monetize properly from user data. To the contrary, it explains the rising appeal for content data.

· Content Safety is vital for video platforms and video understanding is a matter of branding protection. This has recently be pointed out with latest scandals, such as Facebook failures on detecting the Christchurch Mosque shooting video.

Last but not least: submerged with thousands of daily content, video platforms do not quite know anymore why and how to get their hands on content data. For the How part, we would argue at Reminiz that we have created a pretty effective solution. As for the why, this is the purpose of this article. Here are the main stakes of content data management for video platforms and why it is an absolute necessity.


Creating premium revenue out of non-premium content

First things first, let’s assume that all the following reasons serve a major purpose for video platforms : improving revenues. For most part with advertising, for some with payed subscriptions. When it comes to advertising, it is one thing to bet on premium content and ensure quality, timing and exclusivity of delivery. Sure, yesterday’s official video of the last touchdown from the Superbowl will score in terms of advertising revenues. But today, premium content only represents a small fraction of the video ocean. We assumed that videos with less views, poorer quality and not so exclusive content did not represent an amazing market for advertisers. Enters the long tail effect. Mastering content data on your video platform enables creating high value insights on non-premium items. The sum of these small insights, video per video, creates overall way higher value than the big video from your homepage.


Being highly reactive on valuable content

Yesterday’s traffic is gone. And that might be the main reason why content data at scale is key in improving revenues. Except for highly premium content, a video loses its full potential a few hours tops after being published. Multiply that by tens of thousands 24/7 and you’ll get why video understanding at scale is the only solution. Getting detailed insights on the video should be as instantaneous as publishing the video, right because this information will help you anticipate the value of it, before it disappears amongst millions of videos.

Once you made sure you were as reactive as possible, a well-tagged video also serves a reuse strategy on ever-green content, meaning content that, if well labelled, will keep high value through time. The better your content is, the higher the chance it keeps spreading over time.


Keeping control on content on your platform(s)

Ever growing video inventories also mean increasing risks to publish non suitable videos on your own platforms and it doesn’t take much for advertisers to be particularly upset if their appearance does not match their expectations. Brand Safety over years has become central to advertisers and publishers. Nudity, violence, decapitation or mass shootings are obvious reasons but not the only ones. Appearing next to a celebrity involved in a scandal, matching with a brand that has bad press or even just content with values you would not want to carry. User generated content added to the vast number of videos published every day make this filtering job hard. Facebook to name but one has a dedicated team assigned to manually identify and remove such content. But proof is that no one can do without video understanding.

Keeping control on your content also involves staying in legality. Pirate videos, broadcasting pieces without the rights, are also at stake here. Video understanding can help video platforms identify bad students in a quick and effective way. In a nutshell: making sure that what you sell (to advertisers and viewers) isn’t rotten.


Capitalizing on your own Database

Content database first. Identifying high values content profiles from deceiving ones is key in inventory management. Many OTTs offer an infinite catalogue even though 80% of it is almost unwatched. Sure, large inventories are part of a strategy, to make viewers think there are so many content out there. But the truth is that it’s sometimes hard to identify coherent patterns in terms of content likability. Content data is not just about putting a few labels on a video. It’s understanding at scale what gives higher chances for a content to be popular.

Celebrity database second. Video platforms publish so many videos from various sources they often do not even know who they feature. Recent assessments at Reminiz have shown that most valuable videos were featuring “unknown” 2.0 icons that any serious database should be aware of. A job that is sometimes hard to make when you try to spot them amongst millions of content manually.


Improving granular targeting at scale

It all comes to this: finding the 100% verified and adequate video with adequate advertising for the adequate user. So far, video platforms have mainly focused on basic labels to sort their inventories out: is this movie sad or joyful, does it feature a top celebrity or random actors. But true content data management implies digging deep into the core nature of each video and making assumptions on a high quantity of videos. Eventually, it might not just be sorting out sad from joyful videos, but for instance automating the perfect timing to advertise in a sad video featuring second rank celebrities wearing branded piece of clothes. Or ensuring your ad goes on only if the football player of your choice scored a goal in the video. Or even identifying users being fond of jungle movies with water settings and people crying (because why not ?)

Content data at scale is at its very beginning, essentially because video platforms have not yet fully grasped the tremendous value to automate surgical video understanding for all content. Now’s the perfect time to change.

Eventually, data at scale serves two major purpose for video platforms : better recommendation AND better monetization. Higher recommendation involves a qualified audience, videos with more views and a lasting lifetime, hence more potential to advertise on it. RGPD has had an impact on direct advertising revenues and growing inventories an impact on storing costs. Content data at scale is right in the middle of these challenges, helping video platforms to recommend, monetize in real time and multiply revenues on each and every content.

Reminiz Insights

Reminiz is a world pioneer video understanding technology offering real-time facial and logo recognition. Augmented Content for a never-seen viewer experience.

Paul Chaumont

Written by

Product Manager at Reminiz

Reminiz Insights

Reminiz is a world pioneer video understanding technology offering real-time facial and logo recognition. Augmented Content for a never-seen viewer experience.

Welcome to a place where words matter. On Medium, smart voices and original ideas take center stage - with no ads in sight. Watch
Follow all the topics you care about, and we’ll deliver the best stories for you to your homepage and inbox. Explore
Get unlimited access to the best stories on Medium — and support writers while you’re at it. Just $5/month. Upgrade