Table of Content

1. How do these popular OTT platforms earn money?
2. The algorithms below make up the recommendation system Netflix uses.
3. How to approach the Netflix modeling recommendation engine?
4. What kind of data is required?
5. Collaborative filtering
6. Content-based filtering
7. How B2Metric IQ OTT Analytics Platform Use AI For OTT Applications
8.  B2M IQ Analytics Integrated Data Sources

 

OTT stands for “Over The Top” and refers to any streaming service that delivers content over the internet. The service is delivered “over the top” of another platform.

In fact, the term OTT is a short form of over-the-top. When you got the facility to watch TV content on the Internet, that is, when you got rid of the cable box and you could watch all the TV programs on a smart phone in your hand, then this platform was called OTT. Streaming of web series, comedy shows and movies is very popular on TV's OTT platforms.

How do these popular OTT platforms earn money?

The arithmetic of film earnings on OTT platforms is straightforward. OTT has to buy the rights to the movies for release or streaming. The producer gets a sum for the rights. This deal is different for different language versions of the same film, that is, the rights of each version are different.

On the other hand, some films are made by OTT platforms. That is, the OTT platform for a particular film makes a deal. Like Netflix is an OTT platform, which is specifically in the business of making films for its platform. In this deal, it happens that the platform gives a fixed amount to the filmmakers and the producers make the film for less than that, that is the remaining amount is their profit.

Overall, the business model on OTT is very simple. First, the platform spends money to make or buy its content, and then the content is sold by charging a charge from the audience or users. Creating an OTT platform is a huge haptic process and at the same time it is a negative cash flow system in the business world. On the other hand, many platforms also provide payment systems such as per week, per month, daily and per annum for the convenience of the users.

The OTT industry is largely betting on artificial intelligence (AI) and machine learning (ML) to grab 30 crucial seconds of the targeted customer’s attention.

Today, everyone wants an AI equipped online streaming platform that can understand their preferences and taste without merely running on autopilot. And Netflix is leading this battle when compared to other OTT platforms like Amazon Prime and Disney+.

Netflix is the world’s leading Internet television network and the most-valued largest streaming service in the world. The amazing digital success story of Netflix is incomplete without the mention of its recommender systems that focus on personalization.

Netflix uses machine learning and algorithms to help break viewers’ preconceived notions and find shows that they might not have initially chosen.

Machine Learning personalization utilizes algorithms and predictive analytics to dynamically present the most relevant content or experience for each visitor. ML personalization provides a more scalable way to achieve unique experiences for individuals, rather than segments of people. It allows you to utilize algorithms to deliver these one-to-one experiences, typically in recommendations for products or content. With a next-generation platform, you can apply ML personalization to recommending categories, brands, offers, and more.

In simple work, the personalization supported by AI/ML technology can be defined as delivering the right content, to the right user, at the right time, and across all the touchpoints.

The algorithms below make up the recommendation system Netflix uses.

● The personalized video ranker – This algorithm orders the entire catalogue of videos for each member profile in a personalized way. It is responsible for the “genre rows” (i.e. rows like Movies with a Strong Female Lead).

● Top N video ranker – This algorithm produces the recommendations for the “Top picks” row, it finds the best few personalized recommendations in the entire catalogue for each member.

● Trending Now – This algorithm detects the “short-term trends” for the “Trending Now” row. There are two types of trends, yearly trends (such as Halloween, Christmas, etc.) and one-off events that spike an interest in a certain category or movie (such as a hurricane spiking an interest in natural disaster documentaries).

● Continue Watching – This algorithm ranks the videos in the “Continue Watching” row based on how likely you are to resume watching a certain video.

● Video-Video similarity – This algorithm deals with the “Because you watched …” rows. Which is a two-part process; with the first part being the generation of a list of similar videos for each video in the catalogue (which is un-personalized), and the second a personalized ranking of each video within the row.

● Page Generation: Row Selection and Ranking – This algorithm is responsible for generating the whole page (i.e. which rows to show where in the page based on relevance and diversity)

These are not the only algorithms responsible for the personalization. The information on the left of the page, such as the short summary of the video, awards,cast, the thumbnail or other metadata, is all generated by what they call “evidence selection algorithms”.

Search is another one of the personalized features offered by Netflix. Even though it amounts for only 20% of the chosen videos, it requires its own set of algorithms. Since users can search for anything (videos, actors, genres) that might not be in the catalogue, search itself becomes similar to a recommendation problem. The search algorithms combine metadata, search data and play data to arrive to the results. You can search for a movie name, a director, an actor/actress, a genre, the video quality and even the language type. However, it is possible that the Netflix catalogue in the country you are in does not contain the video you are looking for, in that case Netflix will try to recommend similar videos and sort them on which one you would like the most.

All these algorithms and processes come together to provide a good viewership experience. Netflix has done a phenomenal job of applying AI, data science, and machine learning the “right way”. Majority of their users consider recommendations with 80% of the views coming from the service’s recommendations.

How to approach the Netflix modeling recommendation engine?

● Prepare your data. Find out the features of users and products to be recommended. Additionally, collect attributes (likes, views, etc.) that describe users' interactions with products.

● Set the performance metric. Split a test set to compare your model's suggestions to and choose a benchmark metric. Optionally, you can follow third different stages. The first is the training metric (for example RMSE), the second is the ranking metric (for example NCDG), the third is the online metric (A / B test).

● Define recommendation strategy. Different recommendation algorithms give recommendation judgment with different logic. For example, collaborative filtering makes you consume content that similar people do not watch, while content-based filtering recommends content similar to the content you consume. Also, the recommendations ability to be session-based can be important in your choice of context, such as suggesting entertaining content when you are looking for entertaining content, or suggesting dramatic when you are dramatic. In addition, how much you will give importance to past likes, that is, how quickly you will adapt to the change in your content consumption style is also a design parameter

● Choose appropriate model. Choose whether to use collaborative filter, latent factor model or deep neural network, whichever fits your design in step 3.

● Training & Tune & Reiterate

● Track real time metric of your recommender’s performance.

● Schedule retraining interval.

 

 

What kind of data is required?

Entities User: Demographics and details about user profile: Age, Gender, Subscription fee preference.

Product: Infos about product: Title, Genre, Actors

Entity Interaction (User->Product) Data Implicit Feedback: Interactions with service: Viewing history and durations, Clicks, Likes, Bookmarks, Duration of review, Devices used

Explicit Feedback: Enjoyment Feedbacks: Ratings

Deep Neural Nets in Recommendation Systems How to Recommend; Feed user & product informations to network to get interaction strength between user & product. Sort products by interaction strength and recommend Neural network architectures can be used to to model interactions between users and products, embeddings (profiles) of users and products. Latent vector models also can be modified to work with neural network architectures

Collaborative filtering

Collaborative filtering takes into account similarities in taste, meaning that if person X liked Harry Potter, The Lord of the Rings and The Chronicles of Narnia movies but hated Transformers, and person Y likes Harry Potter and The Lord of the Rings movies, we assume he is more likely to like The Chronicles of Narnia than Transformers. It is, thus, based on the similarity between users, not products.

Content-based filtering

Content-based filtering, on the other hand, models the likelihood of liking a certain movie by looking at the features themselves. In the example above, we could sense that X and Y like the fantasy genre, and recommend The Hobbit for both of them. In this example, it is based on the similarity between products, not users.

 

Most companies use a hybrid approach, meaning they combine collaborative and content-based filtering, taking into consideration not only ratings and categories, but also demographic data such as age and gender, and more advanced features such as the text content in a book, for instance. Especially when you don’t have enough data for a specific user, such as ratings, to be able to infer their preferences from that, you can use demographics to estimate it.

Finally, dig deep and you will see that Netflix generated supporting data before making the strategic move forward. Thus, we can state that Netflix is data driven company and companies across industries can all learn a lesson or two from Netflix’s playbook when it comes to deploying AI solutions.

 

     HOW B2METRIC IQ OTT ANALYTICS PLATFORM USE AI FOR OTT APPLICATIONS

 

B2M IQ OTT Analytics platform provides visualization of user event actions and Customer Behavior Predictive Analytics on OTT software (Mobile, Web, TV). B2M IQ OTT Analytics provide service on the predictive side such as Customer Churn Prediction, Customer Micro-Segmentation, Content Scoring, Customer Next Best Action, Propensity Prediction, App Personalization. The B2M IQ OTT Analytics platform provides the services it offers by easily integrating with the platform where the OTT Platform keeps its data. Thanks to this integration, users of OTT platforms can access the results of event actions by simply looking at the B2M IQ OTT Analytics platform and can easily access the reports they need.

General Video Metrics Report

● General Report is lists of total user and view counts, Top 10’s. This report can be customized for the platform needs.

● Categoric Report is the count of recently and all times published contents total users and views.

● Search & Keyword Report is count of search bar keywords and clicks.

● Detail Report is the chart of the all parameters.

Content Metrics Report Summary & Cohort Analysis

● Based on the last day of the specified date range, it gives the number of unique users viewing on a monthly, weekly and daily basis.

Content Metrics Report Summary & Funnel Analysis

● The log-in rates of the users who come in weekly periods within the specified date range are shown.

B2M OTT Applications Predictive Analytics Dashboard

● DAU (Daily Active User): The number of users logged into the application at least once a day.

● DEU (Daily Engaged User): The number of users who watched the video at least once a day.

● WAU (Weekly Active User): The number of users logged into the application at least once a week.

● WEU (Weekly Engaged User): The number of users who watched the video at least once a week.

● MAU (Monthly Active User): The number of users logged into the application at least once a month.

● MEU (Monthly Engaged User): The number of users who watched the video at least once a month.

Categorical Report

● The number of impression, how long a video was watched.

● The number of unique viewers refers to the number of people who watched the video.

● Statistics by categories that you can choose. (E.g. User, view counts, rates…)

Dashboard Search & Keyword Report

 

● The number of clicks on the categories in the time period filtered by categories.

 

 ● The totals of the words searched in the search bar for the filtered time range and all time. 

 

                                   B2M IQ ANALYTICS INTEGRATED DATA SOURCES

 

 

Data Sources are intended to help users and applications connect to and move data to where it needs to be. They gather relevant technical information in one place and hide it so data consumers can focus on processing and identify how to best utilize their data.

The purpose here is to package connection information in a more easily understood and user-friendly format. This makes data sources critical for more easily integratin disparate systems, as they save shareholders from the need to deal with and troubleshoot complex but low-level connection information.

And although this connection information is hidden, it is always accessible when necessary. Additionally, this information is stored in consistent locations and formats which can ease other processes such as migrations or planned system structural changes. 

User Related

● The time users spend on the platform on a daily, weekly, monthly basis. 

● Which categories the user distributes their time, 

● How much of the time spent on the platform does spend watching for content?

Platform Related

● Application download counts of the Android, IOS breakdowns,

● Click counts of the categories,

● Keyword names and frequencies as entered in the search bar.

Platform - Activity Related

● A daily, weekly, monthly basis,

● Singular active user count,

● Singular active user count for the country, city breakdowns,

● Singular active user count of the platform(Android ~ IOS) and device (tablet - mobile) breakdowns;

● The active users count at a certain breakdown range. (For example 1, 5, 30, 60+ minutes.),

● How much of the time spent on the platform does spend watching for content? (Platform Value Time),

● The rates of entering on the platform from the user’s register date to the following days. (Cohort Analysis),

● Average time each user spends on the platform.

● Distribution of the time spent on the platform's episodes(Main Playlist, Explore), etc. of the application.

Dashboard General Report

Customer Lifetime Value Purpose

The customer lifetime value is to assess the financial value of each customer. Customer lifetime value differs from customer profitability in that customer profitability measures the past and customer lifetime value looks forward. While quantifying customer profitability is a matter of carefully reporting and summarizing the results of past activity, quantifying customer lifetime value involves forecasting future activity.

Customer Lifetime Value Advantages

● Management of customer relationship as an asset

● Monitoring the impact of management strategies and marketing investments on the value of customer assets,

● Determination of the optimal level of investments in marketing and sales activities

● Encourages marketers to focus on the long-term value of customers instead of investing resources in acquiring "cheap" customers with low total revenue value

● A natural decision criterion to use in automation of customer relationship management systems

● Implementation of sensitivity analysis in order to determinate getting impact by spending extra money on each customer

● Optimal allocation of limited resources for ongoing marketing activities in order to achieve a maximum return

● A good basis for selecting customers and for decision making regarding customer specific communication strategies

● Measurement of customer loyalty (proportion of purchase, probability of purchase and repurchase, purchase frequency and sequence etc.)

The disadvantages of CLV do not generally stem from CLV modeling per se, but from its incorrect application.

Content Recommendation

AI applications develop for the best user communication on OTT platforms and analyze their actions for them. B2Metric exists to predict user’s next step and make the best option for them.

● Make sense of users’ viewing behavior.

● Recommend the best content to the user and increase your view.

● Analyze user historical data and prediction the next best action.

● Smart marketing strategy with the best suggestions.

● Predictive the next content based on previous trends.

● Forecasting of product demands by users and platforms.

● Create micro segments of customers

Connect & Run Predictive Analytics From All Data Sources

● Seamlessly Google Analytics, BigQuery, AWS Redshift, Adjust, Mixpanel, Countly Analytics Integration and your own DB.

● Easily integrate Auto-ML solution BMS.

● Strengthen your model with external data sources provided by us.

● Analyze user profiles and views with BMS.