This past year has seen a monumental shift in video consumption. One does not need to read reports like this and this to understand the shift towards video and CTV. We have reached a point where video viewing has become a seamless part of our lives. When taking a break from entertainment, we engage ourselves in educational videos. And videos we do watch!
Naturally, advertising dollars have increasingly flown towards this medium. And yet, video advertising seems to be intrusive. Everyone is left wondering the ROI on targeted video advertising. In addition irrelevant video advertising has already started deteriorating the consumption experience and is soon to be in danger of being a nuisance to be ignored.
This is mainly due to not accounting for ad relevance to the video content. With most targeting still dependent on highly inaccurate audience data, we are left with a sinking feeling of being in a dentist chair when interrupted by ads. Even when using context for decisioning; “context” is generally inferred through text associated with the video utilizing pre-defined concepts (auto, garden, etc.)
Video advertising will have to be more contextual by gaining a semantic understanding of the various events in the video through the analysis of video frames, text, speech and associated meta-data. Given the video of a car driving through the streets and meeting with an accident, a semantic understanding would suggest an insurance ad whereas a video of a car driving through the mountains and meadows may suggest an ad for a car manufacturer. We achieved semantic understanding using the following framework -
- Concept Identification : Identifying the subjects and the overall location of the subjects in the video (man in a coffee shop, baseball game in a stadium, etc.)
- Event understanding: Understanding the event associated with the concept (man drinking coffee, home run being hit, etc.)
- Theme understanding: understanding the overall theme of the video (adventure travel, food experience, etc.)
By segmenting the video into meaningful chunks and associating the above framework provides us with a narrative of the video. At Netra we have built our technology with this vision of being able to describe the video context, rather than just identifying objects or providing tags, in a way that this rich medium can then be broken down into its semantic understanding. This can in turn be automated into decisioning, real time decisioning to extract the value of the video asset. This will provide the business value that systems are looking for
The critical part is to be able to scale this to be impactful in programmatic and keeping up with the hours of video content posted online or produced on TV. In addition
This needs to be done in real time for live streams and sports events. At Netra, once we cracked this nut, we have been able to make video advertising relevant again for our partners and the ecosystem!