Explorable visual analytics for media delivery platforms

Ericsson MediaFirst is a next-generation software-defined TV platform for the creation, management and delivery of pay TV content. It is fully optimized for media delivery and offers many advanced features and personalization technologies. The suite comes with deep instrumentation technology allowing live collection of thousands of KPIs – both from the backend systems as well as from the client side. These high-end monitoring functions enable not only fault prevention and diagnostic measures, but also advanced analytics and real-time decision-making practices. The information obtained from these functions can then be used to devise novel marketing strategies to attract new subscribers, upsell and retain existing subscribers, or explain various customer usage patterns.

The sky is the limit as to what can be done with such information, but then we asked ourselves where we must start looking for insights.

This question brought Ericsson Garage and xSeer together to explore how xSeer’s Explorable Visual Analytics (EVA) tool could help to further extend the potential use of the data collected from the Ericsson MediaFirst platform. The technology behind xSeer was developed by Amir Yahyavi and Saman Amirpour as postdoctoral researcher and PhD candidates working at Carnegie Mellon's CREATE Lab. EVA is a powerful big-data visual-analytics platform that can take billions of rows of data with more than 100 columns and present it in a visual format that allows us humans to obtain a better understanding of what is happening. EVA offers in-depth ad-hoc drilldowns into the data over time, data blending, and large-scale machine learning to discover business insights. This in turn suggests courses of action to achieve business objectives. The goal with EVA is to speed up hypothesis generation by easy data exploration and quick intuition validation.

Next, we asked ourselves what EVA could do with the Ericsson MediaFirst data. We explored a couple of interesting patterns under xSeer’s "Navigate to find Insights" paradigm. First, we categorized the proof points into three groupings, notably client Quality of Experience, Business Intelligence, and Operational Support insights. We decided to tackle the first two given the available test data and time restrictions. When it came to client end-point QoE, we used EVA to show a geographical time lapse of various client data through stalls per minute. We then introduced census data information such as age and income groups. We combined this with Time Since Last Visit (TSLV) – a potential customer churn prevention indicator, but what would be valuable for an operator is to know how to actually prevent the churn before it occurs. With xSeer, we showed TSLV and the number of stalls for twelve months for each of a series of blocks across the US.

Firstly, we viewed TSLV over the map. Then, we pin pointed areas where the TSLV increased dramatically compared to other neighborhoods. We then selected that area to verify if there is any sign of correlation between number of stalls and TSLV. We also incorporated census data about age groups. This verified that higher number of stalls and lower age range resulted in larger TSLV values. One action the operator could take to remediate customers from the stall issue – such perform root cause analysis – with more focused attention to clients in younger age groups and perhaps advertise improved quality to those age groups as well.
Another interesting use case we investigated was the popularity of various shows. For each show, we drilled down to areas where they are most popular. We then tried to test various hypotheses using census data such as age, income and education. This information is valuable for direct marketing such as ads or recommender system setting for specific content.

The video below demonstrates these use cases on EVA.