In this project we asked the question “how would consumers find the content they will enjoy watching”. It featured a combination of discovery, recommendations, and personalisation; a mix of lean-forward and lean-back interactions, across 10-foot-UI, mobile, and tablet.
As part of this project we created partnered with a customer of Ericsson Broadcast and Media Services – one of the largest Nordic broadcasters which was consumer-facing and thus had a content library and playout services including broadcast and Video-on-Demand. We created multiple prototypes using their content library and performed user studies.
The prototype that was best received was Tiled Infinite Scrolling. Here, users could scroll down forever, but if they found themselves in a neighbourhood they did not like they could swipe left and be thrown in to a completely different neighbourhood. For example the system could infer that this user enjoys Action movies, and therefore the user finds herself viewing Action movies, and she keeps scrolling to find Action Comedies, followed by Situational Comedies, followed by Romantic Comedies. At this point she decides she is uninterested with what is being recommended so she swipes left and find herself receiving recommendations for Psychological Horror, which is completely different than what she’s seen before. If this is interesting she can scroll down and if not she will swipe left to again find something different from anything already recommended.
This prototype displayed some brilliant visual UI work from RedBee, and powerful clustering and personalisation algorithms developed in-house.
Role
I was the primary UX resource in a US-based, multi-disciplinary team consisting of engineers and data scientists. I first performed heuristic evaluations to ensure the front-end, back-end, and machine learning models provided the best user experience possible. I then designed the studies, including how to evaluate the prototype(s) and the definition of success, and led the pilot studies in-house. I worked with data scientists to design and iteratively improve machine learning algorithms that better fit user expectations and improve user experience. I worked with internal partners from RedBee Creative (then a subsidiary) for visual design and Ericsson Consumer Lab to facilitate user studies in Stockholm and worked with external partners in the US to support development work alongside the rest of the team.
I also synthesised findings from user studies into actionable items; providing feedback to developers, visual designers, and stakeholders, to ensure we had a significantly better product available for the subsequent study.
Other Proof-of-Concepts
There were many other content discovery and recommendation proof-of-concepts (PoC) that were built and tested. The rate-to-replace PoC replaced recommended items with new items based on the ratings supplied. The example-based discovery PoC allowed user to specify the kind of movie they wanted using existing movies as an example, e.g. “I want to watch something more like Batman Begins, but less like Suicide Squad”. The intelligent channel surfing PoC reimagined what channel surfing would look like with Video-on-Demand. Here we threw users into the middle of the content, and allowed them to indicate if the liked what they saw, which would influence the next piece of content recommended.
Impact
All parties involved had different priorities and received different benefits from this engagement. Ericsson Research was able to secure multiple patents while my team used this project to start machine learning initiatives with the work on recommended systems. Our partners RedBee Creative were able to show off their creative chops for media content discovery. Ericsson’s Consumer Lab started a closer collaboration with Ericsson Research; they are now a research area under Ericsson Research. As a company, Ericsson was able to demonstrate technological and creative superiority to our broadcast customer(s), while the customer received multiple innovative ideas; some of which has inspired their mobile and tablet apps. This work has also been used in other parts of Ericsson.
During this activity I also evangelised user testing to managers and groups who have never done it before. I remember one manager looking at version 3 of our PoC and saying “now I understand why we do this”.
Publications
- Incremental Learning for Matrix Factorization in Recommender Systems
by Tong Yu, Ole Mengshoel, Alvin Jude, Eugen Feller, Julien Forgeat, Nimish Radia
IEEE International Conference on Big Data 2016 - Tag-Based, User Directed Media Recommendations
by Vladimir Katardjiev, Alvin Jude Hari Haran
US Patent US20210049211A1
