An OTT (over-the-top) media service is a streaming media service offered directly to viewers via the Internet.
OTT bypasses cable, broadcast, and satellite television platforms, the companies that traditionally act as a controller or distributor of such content.
The main goal of any OTT platform's recommender system is to get the right media in front each of its consumers.
The thumbnail of the media plays a huge role in demystifying this problem statement.
If the thumbnail can capture something compelling for the consumer, then it will act as a gateway into that media by giving him/her some visual evidence for why the media might be good for him/her.
A thumbnail may have either just the title of the media, or the media's actor's face, or a screenshot of an exciting moment from media, or screenshot of a scene from the media that conveys the essence of the same.
Understand this with the given example of the Indian 2020 biographical, period, action, drama, thriller, movie Tanhaji:
1. If a consumer has seen a few Ajay Devgan's movies recently on your platform, the thumbnail of Tanhaji-movie having his face can make the consumer click and watch it.
2. If a consumer has seen a few Saif Ali Khan Pataudi's movies recently on your platform, the thumbnail of Tanhaji-movie having her face can make the consumer click and watch it.
3. If a consumer has seen a few Indian period movies recently on your platform, the thumbnail of Tanhaji-movie, showing the actor Sharad Kelkar as the Indian king Shivaji Maharaj, can make the consumer click and watch it.
4. If a consumer has seen a few action movies recently on your platform, the thumbnail of Tanhaji-movie, showing the classic "Ajay-on-horse vs Saif-in-air" scene can make the consumer click and watch it.
So, now the point/problem is:
How to find the thumbnail that will be compelling enough for the consumer to click?
One solution to this problem is:
Find the single perfect thumbnail using AB, for each media.
A better solution would be:
Find the best thumbnail for each consumer that highlights the aspects of a given media that are specifically relevant to them.
Challenges in showing a different thumbnail to each different customer:
1.
Because only a single thumbnail can be used to represent each media in each place for a given user, the thumbnail selection becomes a chicken-and-egg problem operating in a closed-loop: if a user plays a media it can only come from the thumbnail that we decided to present to that user.
So, the challenge is to understand that when the user played a media, whether he was influenced by the thumbnail and when the user played it regardless of which thumbnail was presented.
2.
The next challenge is to understand the impact of changing thumbnail for a given media between sessions - but continuous changes can also confuse people.
Does changing thumbnail reduce the findability of the media and make it difficult to visually locate the media again?
Does changing the thumbnail itself lead the user to now consider it, which s/he had previously rejected?
Changing thumbnails also lead to an attribution problem, as it becomes unclear which thumbnail made a user in a given media.
3.
The next challenge is of understanding how a given thumbnail performs in relation to other thumbnails we put up the same page or session.
Maybe a bold close-up of the main character works for a media on a page because it stands out compared to the other thumbnail.
But if every media had a similar thumbnail then the page as a whole may not seem as compelling.
Looking at each thumbnail in isolation will not be enough and we need to think about how to select a diverse set of thumbnails across various medias on a page and across a session.
The effectiveness of a thumbnail for a media will also depend on what other types of evidence and assets (e.g. synopses, trailers, etc.) we also display for that media.
4.
To achieve effective personalization, we will need a good pool of thumbnails for each media - By good, we mean engaging, informative and representative of a media to avoid “clickbait”.
The set of thumbnails for a media also needs to be diverse enough to cover a wide potential audience interested in different aspects of the content.
5.
Finally, there are engineering challenges to personalize thumbnails at scale.
When you have a platform having millions of media, and hence millions of thumbnails, using personalized selection for each media means handling a peak of over million requests per second with low latency.
Such a system must be robust: failing to properly render the thumbnail in UI would significantly degrade the experience.
Also, the personalization algorithm also needs to respond quickly when a new media goes LIVE on the platform, which means rapidly learning to personalize in a cold-start situation.
Then, after launch, the algorithm must continuously adapt as the effectiveness of thumbnail may change over time as both the media evolves through its life cycle and member tastes evolve.
OTT bypasses cable, broadcast, and satellite television platforms, the companies that traditionally act as a controller or distributor of such content.
The main goal of any OTT platform's recommender system is to get the right media in front each of its consumers.
The thumbnail of the media plays a huge role in demystifying this problem statement.
If the thumbnail can capture something compelling for the consumer, then it will act as a gateway into that media by giving him/her some visual evidence for why the media might be good for him/her.
A thumbnail may have either just the title of the media, or the media's actor's face, or a screenshot of an exciting moment from media, or screenshot of a scene from the media that conveys the essence of the same.
Understand this with the given example of the Indian 2020 biographical, period, action, drama, thriller, movie Tanhaji:
1. If a consumer has seen a few Ajay Devgan's movies recently on your platform, the thumbnail of Tanhaji-movie having his face can make the consumer click and watch it.
2. If a consumer has seen a few Saif Ali Khan Pataudi's movies recently on your platform, the thumbnail of Tanhaji-movie having her face can make the consumer click and watch it.
3. If a consumer has seen a few Indian period movies recently on your platform, the thumbnail of Tanhaji-movie, showing the actor Sharad Kelkar as the Indian king Shivaji Maharaj, can make the consumer click and watch it.
4. If a consumer has seen a few action movies recently on your platform, the thumbnail of Tanhaji-movie, showing the classic "Ajay-on-horse vs Saif-in-air" scene can make the consumer click and watch it.
How to find the thumbnail that will be compelling enough for the consumer to click?
One solution to this problem is:
Find the single perfect thumbnail using AB, for each media.
A better solution would be:
Find the best thumbnail for each consumer that highlights the aspects of a given media that are specifically relevant to them.
Challenges in showing a different thumbnail to each different customer:
1.
Because only a single thumbnail can be used to represent each media in each place for a given user, the thumbnail selection becomes a chicken-and-egg problem operating in a closed-loop: if a user plays a media it can only come from the thumbnail that we decided to present to that user.
So, the challenge is to understand that when the user played a media, whether he was influenced by the thumbnail and when the user played it regardless of which thumbnail was presented.
2.
The next challenge is to understand the impact of changing thumbnail for a given media between sessions - but continuous changes can also confuse people.
Does changing thumbnail reduce the findability of the media and make it difficult to visually locate the media again?
Does changing the thumbnail itself lead the user to now consider it, which s/he had previously rejected?
Changing thumbnails also lead to an attribution problem, as it becomes unclear which thumbnail made a user in a given media.
3.
The next challenge is of understanding how a given thumbnail performs in relation to other thumbnails we put up the same page or session.
Maybe a bold close-up of the main character works for a media on a page because it stands out compared to the other thumbnail.
But if every media had a similar thumbnail then the page as a whole may not seem as compelling.
Looking at each thumbnail in isolation will not be enough and we need to think about how to select a diverse set of thumbnails across various medias on a page and across a session.
The effectiveness of a thumbnail for a media will also depend on what other types of evidence and assets (e.g. synopses, trailers, etc.) we also display for that media.
4.
To achieve effective personalization, we will need a good pool of thumbnails for each media - By good, we mean engaging, informative and representative of a media to avoid “clickbait”.
The set of thumbnails for a media also needs to be diverse enough to cover a wide potential audience interested in different aspects of the content.
5.
Finally, there are engineering challenges to personalize thumbnails at scale.
When you have a platform having millions of media, and hence millions of thumbnails, using personalized selection for each media means handling a peak of over million requests per second with low latency.
Such a system must be robust: failing to properly render the thumbnail in UI would significantly degrade the experience.
Also, the personalization algorithm also needs to respond quickly when a new media goes LIVE on the platform, which means rapidly learning to personalize in a cold-start situation.
Then, after launch, the algorithm must continuously adapt as the effectiveness of thumbnail may change over time as both the media evolves through its life cycle and member tastes evolve.
Credits:
Microsoft.com/en-us/research/wp-content/uploads/2016/02/p661.pdf
Netflix.com/in/
En.Wikipedia.org/wiki/Over-the-top_media_service
NetflixTechBlog.com/selecting-the-best-artwork-for-videos-through-a-b-testing-f6155c4595f6
En.Wikipedia.org/wiki/Recommender_system
Media.Netflix.com/en/company-blog/the-power-of-a-picture
NNGroup.com/articles/personalization/
NetflixTechBlog.com/artwork-personalization-c589f074ad76
NNGroup.com/articles/customization-personalization/
No comments:
Post a Comment