Netflix Recommendations and Visual Personalization
With binge-watching audience behaviour on the rise, there are many of us who are all too familiar with the UX of the video platform.
Up until online video sharing and platforms, audiences were stuck watching whatever movie or show that their cable provider had decided to show them. The internet however has led to the creation of a high choice environment where people have more options than ever. This high choice environment however brings with it information anxiety and decision fatigue.
Decision fatigue in the simplest of terms may be described as the phenomenon of having a full wardrobe and nothing to wear or perhaps have 100s of channels but nothing to watch. To curtail this decision fatigue and in an effort to curate the information overload that occurs on the internet, were developed customization and targeting tools.
Netflix in that regard is notorious for its targeting technology. People are always baffled by the recommendations that they often see targeted towards them. From the first instance of use to regular use (Netflix, 2019). Sometimes it seems that the program hits the nail right on the head and other times not so much.
What’s interesting about Netflix however is that their targeting technologies go beyond surface-level user behaviour. According to the Netflix blog on Medium, targeting or recommendations have an added layer of what artwork will best represent the content as well as appeal to the audience member they want to compel to click on it (Chandrashekar et al, 2017).
Traditionally, you’d create user personas based on gender, location, socio-economic class, habits, preferences and like-minded audiences. It was very likely that two 20-year-old male users would come across the same advertisements if they had similar preferences. Netflix takes it one level further by taking their aesthetic preferences into account. Below is an example of Stranger Things thumbnails as taken from the Netflix Blog (Yu, 2019).
With these 9 images serving as a Rolodex, users will be shown certain images based on their aesthetic preferences. These preferences are based on visuals that they may have clicked on or hovered on previously.
If you look at the above images, what’s most interesting is that it takes into account the brand identity of Netflix, the aesthetic language/branding of the show itself and then customizes based on visuals and placement or size of text only.
In this experiment, the Netflix recommendations of three users will similar personas will be compared with one another to see what the aesthetic recommendations are for each.
What goes into creating a good Netflix Thumbnail?
According to Netflix, one of the biggest motivating factors for a higher click-through rate (CTR) was that of favourite actors. Another aspect was that of genres. If there’s a superhero movie
The four elements that will be compared will include
- The Home Page
- A Netflix Original production (The Good Place)
- An old TV show (Friends)
- A made for digital Documentary (Making a Murderer)
Personas
Below is a snapshot of the personas of the participants in this experiment.
Comparison
Home Page
Below is a representation of the home pages for the above 4 personas. As we can see, the recommendations vary from mobile to desktop and so do the artworks for the thumbnails. For User H who is a PhD student, Netflix recommends the Witcher while for the remaining 3 users. The recommendations are for the new season of ‘You’. What’s interesting is that for User ‘H’ the thumbnail for the ‘Witcher’ is that of its lead female character, who is not the protagonist of the show. Even on Mobile, the main lead does not stand alone. This could be an indication that the user tends to watch shows with strong female leads. Another point of interest is that User C’s PhD research is within the realms of feminism and the concept of the ‘other’. For the show ‘You’, the thumbnails are usually standard except for User M on desktop. User R and C are shown a hint of a love triangle while User c is shown a more dramatic visual with fire in the background and a more strained relationship between the two characters.
Netflix Original Production — The Good Place
For a Netflix original production, User C and user R once again have standard thumbnails, both for desktop and for Mobile whereas User H has an image with more of a group and 2 of the female characters in the show. User M once again has a more dramatic image with the male lead of the show without a shirt. This shows that perhaps User M tends to click on visuals that are of a stronger nature whether sexual or violent.
Another thing to note is that the search recommendations yield similar shows along with the requested one and while most of the recommended shows are uniform across users, User H is shown a different one at search result 5 and 6 and once again has thumbnails that show female leads rather than male ones such as in the case of ‘Atypical’.
Old TV Show — Friends
The purpose of choosing an older TV show is to check whether the limitation of not having high-resolution artwork not specifically designed to Netflix guidelines is a hindrance. This is to check whether older shows have more standardized thumbnails as compared to newer ones.
For mobile, all users had the same thumbnail of Ross wearing a t-shirt whereas the thumbnails for desktop differed slightly. User H was once again targeted with a female protagonist with User M and Users R and C were shown a picture of Chandler. This could simply be an indication of which character or actor these users preferred. User R confirmed that the character shown ‘Chandler’ was in fact her favourite on the show.
Made for Digital Documentary — Making a Murderer
For Making a Murderer, we can see that the images once again are rather standard across Mobile. User M was once again shown an image that is more striking and the colour is more eye-catching. What’s interesting is that this is also the thumbnail that is used across all users when it comes to desktop. Perhaps this could just be a result of the desktop ecosystem which lends itself to more clutter than mobile. As we can see, we see 3 results per row on Mobile and 5 per row on desktop.
Discussion
It is rather evident that users are in fact shown differing artworks based on viewing habits. One thing that was also of interest that gender did not play a role when it comes to artwork creation. With one Male participant to see how much the visuals would differ, we found that they were in fact more prone to see a more standard image. In fact, the most diverse and more stereotypically dramatic images were shown to a female participant ‘User M’. If we consider their viewing habits as shown in the persona table above, we can see that the content they consume (Crime dramas and documentaries) is in fact of a dramatic nature.
References
Blattman, J. (2018). Netflix: Binging on the Algorithm. Medium. Accessed 19th December, 2019. Available at < https://uxplanet.org/netflix-binging-on-the-algorithm-a3a74a6c1f59>
Chandrashekar, A., Amat, F., Basilico, J and Jebara, T. (2017). Artwork Personalization at Netflix. Medium. Accessed 19th December, 2019. Available at < https://medium.com/netflix-techblog/artwork-personalization-c589f074ad76>
Netflix, 2019. How Netflix’s Recommendation System Works. Netflix. Accessed 19th December, 2019. Available at < https://help.netflix.com/en/node/100639>
Plummer, L. (2017). This Is How Netflix’s Top-Secret Recommendation System Works. Wired. Accessed 19th December, 2019. Available at < https://www.wired.co.uk/article/how-do-netflixs-algorithms-work-machine-learning-helps-to-predict-what-viewers-will-like>
Yu, A. 2019. How Netflix Uses AI, Data Science, and Machine Learning — From a Product Perspective. Medium. Accessed 19th December, 2019. Available at < https://becominghuman.ai/how-netflix-uses-ai-and-machine-learning-a087614630fe>
