Netflix

All data in the world is doubling every two years, and by 2020 we will reach 44 zettabytes of data (i.e. 44 trillion gigabytes). But it is not the amount of data that is important but how we look at it and use it efficiently and how organizations use it for their benefit. This data can be analyzed and used to not only make calculated decisions but also to help to improve the experience of using an application or service.

Netflix is a provider of internet streaming of media. In layman terms, Netflix provides users with movies and TV shows online for viewing. It was founded in California as a DVD rental service in the United States of America but later evolved into this online service available almost all over the world. Netflix is a data driven company. At present, Netflix has 81.5 million subscribers all over the world. The central concept of Internet television is freedom of choice – what to watch, when to watch and why. But humans are impatient and surprisingly bad at making decisions when provided with a plethora of options and that is when a practical and carefully curated recommended system steps in.

Netflix

This system consists of many algorithms that are beautifully arranged to work together to create a masterpiece. This masterpiece is not only capable of predicting what you would want to watch, but also presenting it in a dynamic way that changes concerning conditions and time.

According to various studies, a typical customer loses interest and quits searching after 30-90 seconds of surfing and after scanning through approximately 10-20 titles out of which they read only about three carefully. Hence the whole game is played within this short period, and what helps Netflix win every time? The recommendation algorithm.

Earlier this algorithm was dependent on the number of stars given to movies or TV shows since the company only dealt with delivering DVDs home. But ever since the content is streamed, vast amounts of data is available. The data is so descriptive that it tells the company all about what the user watches and how. It also tracks what device a user uses, the trends in days of the week, the intensity of watching, and also recommendations that were shown but not selected. This data hence indicates that there is more to recommendation than the highest rated video.

The Netflix homepage window is made up of various components of proposals, each of which is brought together by data analysis.
The Personalized Video Ranker (PVR) provides the videos that appear in a row according to the genre. This algorithm orders all the videos in the catalog in a personalized way for every user. Hence when different users see different genre rows, the reason is PVR. It is widely used and extremely generalized hence the degree of personalization is limited and is blended with general popularity.

The next row, “Top Picks” is formulated by the Top N-Video Ranker. The goal of this algorithm is to show those videos that will not only be liked by the viewer but are also popular. PVR uses the entire catalog for its ranking, but Top-N Video Ranker only focusses on the head of the archive containing the top ranked videos and selects the intersection with the choices of the user. It relates popularity with personalization. This algorithm hence identifies and uses view trends ranging from one day to as long as one year.

The “Trending Now” row stresses on short termed, fast paced trends that last a few minutes or hours. This trend when mixed with personalisation forms this row. One of the most important rows on the page, the “Trending Now” row deals with two kinds of patterns: the first being yearly which primarily comprises of festivals and special days that occur annually. For example,
on Valentine’s Day, this row will have more romantic movies made by the user’s favored director or actor. The second pattern involves very short-term events which are happening currently like the increase in viewership of documentaries during the election time.

The “Continue Watching” row deals with not only episodic content which is watched in pieces but also movies that were abandoned by the viewer in the middle. The algorithm sorts the recently viewed videos on the basis of whether the user intends to resume or re-watch the video. The signals it uses are the time elapsed, time of abandonment and intensity of watching.

The “Video-Video Similarity” algorithm puts together the “Because You Watched” row. This algorithm sticks to one video watched by the user and throws suggestions based on that. Even though this algorithm is not personalized, all the suggestions do not get displayed. The ones that do are according to the user’s interest and what they would enjoy.

Finally, the Page Generation algorithm acts like the glue that puts together all these components into one great masterpiece that is the Netflix homepage. Each row is the best estimate of the choices of a unique user. This algorithm thus uses the output of all the others to construct the page. It considers the relevance and importance of each row for the user and since every member will have a significant number of relevant options this algorithm uses a rule-based approach for making the page. Since this algorithm does not conform to any particular design or template, it is fully personalized and dependent on the user’s choices and trends observed by it.

Thus, using data intelligently will not only help the company predict future trends but also help it increase business by providing access to the way a user thinks. This observation is very elementary, but its implementation has a beautiful outcome. We are now successful in creating so much data that, it can become an all new standard of problem-solving. The real-time access to a user’s behavior creates data that is an asset for the company. If used correctly can do wonders.

1 COMMENT

LEAVE A REPLY

Please enter your comment!
Please enter your name here