Netflix Case Study: Data Analytics for Organisational Decision-Making

Question

Task: Task Summary
Using the concepts covered in the modules, write a 2000 words Netflix case study report for a real scenario faced by an organisation of your choice.

Context
This assessment gives you the opportunity to demonstrate your understanding of concepts covered in modules including Business Intelligence, Big Data, Business Analytics, Data Warehousing, Data Mining, AI, Machine Learning. In doing so, you are required to select an organisation and then analyse and evaluate how the above-mentioned concepts can be used to solve a real-life problem.

Task Instructions
Select an organisation that you would like to investigate. When choosing the organisation, make sure that you are able to access data from the organisation easily, or the data is available on the web.
Write a case study report outlining how the selected organisation has used the concepts covered in modules to successfully solve a problem faced by the organisation.

Answer

Introduction
The present report explores the Netflix case study in detail. Netflix is among the world’s largest streaming and production house. However, they are also famous for using advanced machine learning, artificial intelligence and other techniques involving analytics, Data Science, and Machine Learning are all examples of data. It's all a matter of science when it comes to getting people to click on the movies on user’s account. When it comes to what their users are interested in, they know just what to put in front of them. They give movie suggestions based on previous films, but they also give visual suggestions based on previous films (Kumar, et al., 2018).

Reason behind the current choice
Netflix employs big data analytics to better understand its client demographics. With this information, businesses can deliver better service or products to their customers. From a wide range of subscribers, Netflix amasses a massive quantity of data. Data such as location, material seen, user interests, data searched for, and time spent watching are all collected by the app (Turner, 2018). A tailored suggestion based on the user's interest is generated using these factors. Netflix's business model has evolved from Video rental store to video-on-demand, as well as the company is presently developing original series. Businesses such as Facebook, Ebay, and Netflix employ these to generate smart playlists for its customers.

Business of the organization
Netflix had already gone from DVD rentals to international simulcasting throughout the short span of time. Video Streaming, unlike pay Tv, is primarily a matter of personal choice for the viewer. Web TV, on the other hand, offers a large selection of programmes from a range of genres as well as demographics. Netflix is currently producing a variety of movies and television shows (Turner, 2018) and given those massive content repositories, data is extremely critical to the company's success. It's a meltingpot for the internet and narrative in the Netflix universe. Their principal source of revenue is subscription fees from customers. Customers may stream content from a variety of movies and television shows on a variety of web services.

Concepts involved

• Business intelligence - There are a broad number of techniques that may be used to quickly and easily retrieve information about with an organization's current condition based on the best available data. BI solutions and technologies are used to transform data into usable information and assist a company in making informed strategic decisions. BI tools retrieve and analyse large datasets and present findings in graphs, maps, tables, infographics, spotlights and visuals in order to provide users with precise information about the company's current status (Chandrashekar et al., 2017). When it comes to solving business-related issues that are beyond the capacity of humans, "business intelligence" refers to a broad variety of methodologies and concepts. On the other hand, a business intelligence expert must be well in the tools, techniques, and technologies utilised to acquire and evaluate business data. Additionally, persons in this role must have analytical skills to apply corporate intelligence to solve challenges.

• Artificial Intelligence - It is in the area of artificial intelligence (AI) wherein machines with intelligence are being developed. A wide range of tools, processes, approaches, including applications are used to enable artificial intelligence (AI). Systems that replicate human intellect are created via use of computing in the form of artificial intelligence (AI). For example, Al's section briefly reviews include nlp, object recognition (including speech recognition), optimization techniques, and neural networks.

• Machine Learning - In the field of artificial intelligence, machine learning is indeed a subcategory that is used to make decisions and predictions without specifying the programming. Machine learning is based on computers' ability to learn from data. There are several applications for machine learning in computing, genetic decoding, exploratory astronomy, finance and insurance; emotion classifications; telecommunication; medical diagnostics; financial industry analysis; sequencing (Mishra, 2020).

Sources for the data
There were several sites, documents, journals, and publications that had been used to acquire the information for the current case. The research relies mainly on these materials to establish the legitimacy and veracity of the organisation. All knowledge and analysis were accessible for the report, making it more comprehensive and informative. Using these sources has resulted in significant savings in terms of both time and money. A large volume of data is required for any of these sources to provide fresh and useful information (Muzumdar, 2014).

Original problem current organization faced
To accurately predict how many stars people would give a DVD, which might also range from 1 to 5, was a challenging task That was all they could get from a client who've seen the movie, so they concentrated on that one behaviour. The audience's perspective, along with their own statistics and remarks, would not have been able to be measured as a result of this. With the advent of Netflix as a streaming service, the company had access to a wealth of data about its users. This covers the time during the day, what day of the week, and also the amount of times they've viewed (Turner, 2018). It became a major data collecting effort as Netflix's subscription and watching base grew (Dizon, 2018). Netflix relies heavily on user recommendations to choose what to show users. How could Netflix be made more user-specific? That's what they're asking! Netflix is determined to provide a response, regardless of the fact that it is only one question. Products and services are heavily promoted on their site.

Solution implementation
Netflix came up with a complicated algorithm to overcome this problem. Netflix made significant investments on equipment and software in order to build up a recommender system and perform large-scale data analysis. Before proposing a plan of action, it consists of three stages.
1. In contrast to real-time analytics, offline computing is used to analyze data that is not directly related to the user's actions.
2. Efforts to finish are cut.
3. When training this algorithm, no specified amount of data is used at any particular time.

In order to stay current, it must be reprogrammed on a regular basis. For example, model training plus batch computations are done offline. Due to the obvious volume of data these deal with, this would be beneficial to execute them on Hadoop utilizing Pig or Hive. Perhaps there is a wide range of databases supporting the results, not only HDFS. This is why Netflix launched Hermes, an in-house tool. Several of the additional features of Hermes, a publication architecture similar Kafka, include multi-DC support, a monitoring mechanism, thereafter JSON into Avro conversion, and a GUI named Hermes console. Companies need a system that could detect, alert, and correct problems in real time. Netflix's nearline layer is made up of offline statistics and other findings. In addition to EVCache, Cassandra, and MySQL are used. The most essential question isn't whether the data should be preserved, and also how to store it in a far more efficient manner (Lamkhede & Das, 2019). For its real-time event flow, Netflix depends on a system called Manhattan that was developed in-house by Netflix. Similar to Storm, but geared to a specific set of demands compared to the Twitter Storm. Hadoop is being used to regulate the data flow by allowing Chukwa to log in. In order to meet its hardware requirements, Netflix relies significantly on Amazon Web Services. To illustrate this point, they use EC2 instances, which are highly scalable as well as almost fault-tolerant. AWS serves as the underlying foundation for their entire infrastructure inside the cloud (Netflix Technology Blog, 2018).

So, their recommendation engine incorporates techniques such as topic modelling & denoising in order to make the best possible recommendations (Chandrashekar et al., 2017). Matrix factorization, factorization machines, Singular Value Decomposition and probabilistic-based models are all examples of techniques that may be easily modified to tackle a wide range of problems. There were almost 100 algorithms used to predict one outcome. Matrix factorization, singular value decomposition and finally Restricted Boltzman Machines really are just a few of the useful mathematical methods employed (Arthanari & Baskaran, 2018).

Implementation issues
According to a Netflix Developers report, the aforementioned issues were found in regard to discovery service-related recommendations:
• There isn't a movie to demonstrate you that a recommendation system can locate.
• Ensuring any unavailable entities are located, notified, and replaced as necessary.
• Netflix has a difficult time figuring out what the customer is looking for because the search criteria are so limited (Netflix Technology Blog, 2018).
• When a visitors click on a search result, it might be difficult to give quality results instantly.
• Customer satisfaction is enhanced by allowing for a variety of indexing methods and metrics to be used.

Benefits realized to the organization
Netflix's overall user engagement has increased as a result of the deployment of recommendation technologies. This adjustment resulted in a decrease in cancellations and an increase in streaming hours. The low monthly churn of this service is largely due to payment gateway issues, rather than users' deliberate efforts to cancel their memberships. Netflix's personalization and recommendations save up to $1 billion a year. The development and improvement of the suggestions state led directly to an improvement in customer satisfaction. The algorithm that won the Netflix Prize was able to improve the predicted ratings of 'Cinematch', the name given by Netflix by over 10 %. Singular Value Decomposition as well as Restricted Boltzmann Models both diminished RMSE by 89 % and 90 %, respectfully (Netflix Technology Blog, 2018).

Conclusion
It is possible to analyze terabytes of data using advanced analytics technologies. It's all about clever big data whenever it pertains to streaming services such as Netflix. Netflix's success is now so dependent on data and analytics which the firm may be better defined as just a data company instead of one that produces films or television shows. They are an excellent illustration about how data analytics could be utilised to make smart decisions and enhance roi.

Recommendations
• Ensembling methods are effective. Instead than focusing on a single method, numerous methods were used to forecast a single outcome. Individually adjusting and training models does not get the best outcomes. With a proper trade-off involving variety and accuracy, even best outcomes may be achieved. There's been a lot of open studies in the context of collaborative filtering, therefore contests like the Netflix Award can help to promote these open ideas and theories (Jeyasekar et al., 2016).
• Steps and procedures for doing A/B testing can be enhanced by using circumstances rather than algorithms. Reinforcement methods can be used instead of typical recommendation systems to make recommendations to users. The reward may be user happiness, the state could become the current content, as well as the action can be to propose the next quality content to the user.

References
Arthanari, J., & Baskaran, R. (2018). Enhancement of video streaming analysis using cluster-computing framework. Cluster Computing. https://doi.org/10.1007/s10586-018-2284-y
Chandrashekar, A., Jebara, T., Basilico, J., & Amat, F. (2017, December 7). Artwork Personalization at Netflix. NetflixTechBlog; Netflix Inc. https://netflixtechblog.com/artwork-personalization-c589f074ad76?source
Dizon, G. (2018). Netflix and L2 learning: A case study. The EuroCALL Review, 26(2), 30. https://doi.org/10.4995/eurocall.2018.9080 Jeyasekar, A., Akshay, K., & Karan. (2016). Collaborative Filtering using Euclidean Distance in Recommendation Engine. Indian Journal of Science and Technology, 9(37). https://doi.org/10.17485/ijst/2016/v9i37/102074
Lamkhede, S., & Das, S. (2019). Challenges in Search on Streaming Services. Proceedings of the 42nd International ACM SIGIR Conference on Research and Development in Information Retrieval. https://doi.org/10.1145/3331184.3331440
Muzumdar, P. (2014). From Streaming Vendor to Production House: Netflix SWOT Analysis. Netflix case studySSRN Electronic Journal. https://doi.org/10.2139/ssrn.2377151
Mishra, D. K. (2020). CHALLENGES IN TEXT MINING FOR BUSINESS INTELLIGENCE. International Journal of Engineering Technologies and Management Research, 5(2), 301–304. https://doi.org/10.29121/ijetmr.v5.i2.2018.660
Netflix Technology Blog. (2018, March 27). Data Science and the Art of Producing Entertainment at Netflix. Medium; Netflix. https://netflixtechblog.com/studio-production-data-science-646ee2cc21a1
Turner, G. (2018). Netflix and the Reconfiguration of the Australian Television Market. Media Industries Journal, 5(2). https://doi.org/10.3998/mij.15031809.0005.208