From Managing IPL Traffic To Personalising Shows, Here’s Why ML Is The Real All-Rounder At Hotstar

Hotstar has been gaining momentum as the largest online streaming platform hosting movies, television shows, original series, and most importantly, the Indian Premier League. They are streaming the IPL live in six languages — Tamil, Telugu, Bengali, Kannada, Hindi and English. The 60-day tournament is slated to grab more than 150 million people on their mobile, desktop and other platforms. While reasons like long office hours, increasing commute time and overall changes in the social dynamics could be attributed to the growing traction for over-the-top media services platforms like Hotstar, but there is also an intensive technological play that drives, manages and offers personalised choice to the viewers.

Intelligent Platform By Akamai Technologies

While Hotstar has immense popularity among Indian viewers, for the VIVO IPL 2018, Hotstar leveraged Akamai Technologies, the world’s largest and most trusted cloud delivery platform to carry their global streaming. Used for the match between Chennai Super Kings (CSK) and Kolkata Knight Riders (KKR) on 10 April 2018, the viewership for the game peaked at 5.5 million concurrent viewers. This was reported by the company to be the highest number for any single event streamed online by a broadcaster. In 2017 Akamai platform had 4.8 million peak concurrent users. The Intelligent Platform by Akamai saw more than 75 percent of viewers streaming the match from mobile networks. This cloud delivery platform can deliver 95 exabytes of data, ingest 2.5 exabytes of data every year, and interact with more than 1.3 billion client devices per day. This data powers the machine learning engines which automatically improve reliability, performance and security of the overall digital experience. It seamlessly integrates with web and mobile performance, cloud security, enterprise access, video delivery, analytics and reporting solutions to help in delivering superior experiences. Ajit Mohan, CEO of Hotstar said, “We are seeing a dramatic growth in our viewership every month, including in cricket. As data costs fall dramatically, and users increasingly look to their mobile as the primary screen, we are starting to see the emergence of Hotstar as the primary destination in cities with more than a million in population. This trend will accelerate in the next few months, especially during IPL, and we rely on Akamai’s solutions to ensure that we are able to scale in line with this vertical growth in demand”.

Dealing With The Traffic Spikes

Despite all the preparation for hosting live matches and a swarm of online viewers, Hotstar team believes that real time traffic patterns can be very unpredictable. Being a home to live cricket and having done cricketing events in the past, Hotstar depends on older patterns to make decisions. “The ensuing spike in traffic would overwhelm our back-end systems and occasionally disrupt us from rendering video”, Akash Saxena, head of technology, VP Engineering at Hotstar, wrote on the company blog. If  the legacy backend fails to scale beyond a point, it makes them difficult to accept newer customers on the platform. He shared some of the lessons learnt during the journey and some of the techniques they adhere to while handling large traffic on the website. Auto scaling is one such method that the company believes may not always prove to be a useful strategy. “We use auto-scaling to ensure that the right amount of servers always exist in a pool, but we’re always scaled up for the peak”, he said. Saxena also added that clients should not completely rely on the server systems to make decisions. “Clients must be smart about inferring when things don’t look right”, he said. At Hotstar they believe that caching, exponential backoffs and panic protocols all come together to ensure a seamless customer experience. It is important that the cloud-based load balancers be specifically ‘warmed up’ to receive such heavy traffic. The other measures are that cloud infrastructure needs to ensure that appropriate types of instances are available if the fleet starts to scale up, establishing clear thresholds, among others. “At Hotstar, we’re working on ensuring that our platform can scale out to tens of millions of customers and this is just not possible beyond a point with monolithic architectures”, Saxena said. Saxena further explained that their platform has three core pillars:

1. Subscription engine
2. Metadata engine
3. Streaming infrastructure

Each of these have unique scale needs and are tweaked separately. “We built pessimistic traffic models for each of these basis which we came up with ladders that controlled server farms depending on the estimated concurrency”, he said, adding that it is important to know the key pillars and the kind of pattern they are going to experience. Bringing down the latency numbers is also very important. Having lower latencies and smart use of player control can provide smooth viewing experience to customers thus helping with the seamless traffic patterns. It results in fewer customers repeating the funnel, thereby making the process streamlined.

Personalisation And Data Democratisation Is The Key

While high traffic is one of they key areas that OTT platforms like Hotstar deals with, one of the key focus areas is customer experience. With over 100 million downloads, Hotstar is building ML algorithms to derive user intelligence from the raw stream of terabytes of daily data that they collect. ML is primarily used to provide a personalised experience to the user in every aspect such as content feed or the ads they see. The main metric that they follow is the total watch time per user per month. The recommendation engine then learns user preferences from their watch history to derive algorithms specific to the user and the content that he or she wishes to watch. They also use ML models to deal with diversity of users across regions.   Technology also gives them an advantage to generate personalised ads, creating a win-win situation for both consumers and advertisers. “There is continuous innovation in improving our modeling strategies, making better features, be it through latent semantic index, clustering or deep learning among others”, said Amit Sachan, machine learning expert at Hotstar. Hotstar also uses its real-time stream data platform, Knol, which is built to be a single platform for any kind of data exchange at Hotstar. As the app collects close to 10TB data every day including ad impressions, behavioural clickstream data, customer support data, among others, it is used to solve a plethora of business problems. Hotstar is also exploring Word2vec to define similarities between movies and offer better options to its viewers based on their choice.

Other Avenues Where Hotstar Is Making Use of Machine Learning

ML was recently used by the Star group-owned app in the Gujarat and Himachal Pradesh State Election display campaigns by Hotstar. Powered by the end-to-end digital solutions agency Performics.Resultrix, it was used to identify the audiences, and the ads were delivered programmatically, cutting down the manpower costs. Analytics played a big role in identifying viewer profiles and interests, leading to a phenomenal increase in time spent on Hotstar app. The whole campaign was strategised and executed through a mix of automation, ML, data and analytics, and predictive marketing. Hotstar had earlier tied up with Zappr Media Labs for mobile audience analytics. A deal was signed to create a deep understanding of the mobile audience that could be used by Hotstar to create personalised communication through advertising offers.