Netflix Algorithm

From SI410
Jump to: navigation, search

Netflix, Inc. is an American subscription video-on-demand streaming service and production company based in Los Gatos, California. Netflix has over 220 million customers and offers hundreds of movies, animations, and TV shows.[1] Netflix users value this platform for its wide offer, high-quality subtitles, and, personalized recommendations to help users find shows and movies of interest to them, which are at the core of the product.[2] To this end, Netflix develops and uses a recommendation system using algorithms. Netflix lets users choose a few titles that they like when they first make a profile. [3] Based on this, first form a recommendation, and develop the algorithm by tracking the following while you are using the app. Behind this advanced machine-learning-based recommendation system that analyzes user preferences to suggest new series or movies,[4] ethics issues such as user privacy and user control exist because users' data is collected, analyzed, and shared with third parties. The ethics of Netflix's algorithm are analyzed below according to the criteria established by each of the three institutions: Microsoft, Common Sense Media, and Federal Trade Commission, along with many other articles.

"Netflix" by Netflix. Copyright Netflix.

Netflix Algorithm Recommendation Engines

According to News 10, 47% of North Americans use Netflix over other video streaming services. It’s so accurate that 80% of Netflix viewer activity is driven by personalized recommendations from the engine[5]. As of 2016, Netflix's recommended engine saves more than $1 billion a year in customer acquisition costs for Netflix[6]. Its algorithm ranks and constructs titles using machine learning systems such as reinforcement learning, matrix factorization, and other algorithmic approaches. Netflix dynamically gets more and more users (almost 222 million subscribers as of Q4 2021), which also shows that they are constantly understanding peoples’ needs concerning entertainment by its algorithm.[7]

How Netflix started using algorithms

When Netflix first launched in 1997, it was a movie rental service that let users order movies and get them via mail. Since then, Netflix accumulated data on people, and which are big data that include people's information, such as gender, region, age, viewing time, and taste. [8]. Netflix began utilizing data science and analytics techniques in 2000 to suggest movies for customers to rent. As Netflix began to focus more on providing video streaming services, it started to change its recommendation engine to confirm to new criteria and objectives.[9] The organization saw a rising requirement to offer and recommend interesting books to various user categories as it expanded into more client segments and geographical areas. About three percent of the company's 2014 sales, or $150 million, went into building a 300-person team tasked with enhancing the recommendation engine.[10] According to consumer research, the average Netflix user starts to lose interest in the service after exploring and considering potential films for 60 to 90 seconds.[11] It's conceivable that a user will migrate to another streaming provider if they lose interest, and which is something that the business should prevent.

How Netflix’s algorithms work

In essence, Netflix Recommendation Engine (NRE) is made up of algorithms that filter content based on each individual user profile. The engine helps users to find a show or movie to enjoy with minimal effort and spend more time on it. To make it more effective, A total of 1,300 recommendation clusters have been created by Netflix based on user viewing preferences. Finding a movie or TV show that each user will like and doing so as rapidly as feasible are the objectives (estimations are they have just 90 seconds for that).[12]

User starts off creating their profile by selecting several titles that you like [13]. If users do not want to do it, diverse and popular sets of titles are represented for you. If the user makes a few choices, it’s a starting point for the recommendation algorithm to start working. The NRE learns your habits and tastes as you view more works. The result made by the NRE is getting increasingly accurate. It also emphasizes each user's most recent selections[14].

"Netflix recommendation workflow" by Saniya Parveez, Roberto Iriondo. Copyright Towards AI Editorial Team.

By allowing users to give titles a thumbs up or down, Netflix gathers explicit feedback information from them. Although on a bigger scale, this is comparable to the explicit data the corporation gathered during its DVD rental period. Additionally, the business gathers implicit data, which is comprised of user behavioral data that is gathered in real-time when a user explores and interacts with content on the service. [15]. Below are the details.

- User’s interactions with its service; viewing history, searches and rating other titles

- Other members' choices with similar tastes and preferences on our service

- Preference and frequency about the titles such as genre, categories, actors, year of release, etc.

- The time-of-day user watches

- How long user watch, when the user paused, rewound, or fast-forwarded

- What scenes users have viewed repeatedly

- The device used to stream

- Demographic information, such as age or gender, is not included when making decisions

Therefore, Netflix Recommendation Engine represents a ranking of titles in a way that is designed to present the best possible ordering of titles that you may enjoy. The most strongly recommended titles start on the left of each row and go right. This algorithm functions when users try to make searching. The top results will also be personalized[16].


Ethics in Algorithms: Using artificial intelligence and algorithms

The use of Algorithms – to make predictions, recommendations, or decisions has enormous potential to improve welfare and productivity. However, personal information gathered for use in algorithms in this way creates ethical problems in the process of collecting data, sharing it with third parties, and creating new things based on this data. It also implies risks such as unfair or discriminatory outcomes or perpetuating existing socioeconomic gaps.[17]

Nearly early every week, a new report of algorithmic misbehaviors emerges. Federal Trade Commission of the United States government offers important lessons about how companies can manage the consumer protection risks of its algorithms and Artificial intelligence.[18] Companies or any organizations that use algorithms or benefit from algorithms should inform customers transparently how the automated tool is used. Customers should also be notified of the adverse action caused by its use.[19] The company is also encouraged to notify customers whenever they make changes in decisions. In order to do that companies must know what data is used in their model and how that data is used to arrive at a decision. Next, you need to ensure that your decisions are fair by giving consumers access and an opportunity to correct information used to make decisions about them and by ensuring that your data and model are robust and empirically sound. Lastly, hold yourself accountable for compliance, ethics, fairness, and non-discrimination[20].


Privacy

Risk of User Data Leakage and Privacy

Netflix was sued in 2009 for releasing movie rating data of 500,000 subscribers that were identified by Netflix's unique ID number. At the time, Netflix released this "anonymous" information to the public as a resource to develop a contest for a better recommendation algorithm. The public, who was researchers were able to identify subscribers by looking at what ratings they gave movies while they watched the content with those on the IMDB website. This website is an online database of information about movies, television shows, music videos, podcasts, home videos, video games, and online streaming content, including cast, production crew, and personal biographies, plot summaries, trivia, ratings, and fan and critical reviews where people use their real names.[21] Netflix believed it was doing enough to protect user privacy, but that wasn't the case. At last, the lawsuit was settled in 2010. [22] The Netflix case describes that a person's distinctive characteristics, such as names and addresses, can lead to the identification of even some of the information that does not appear malicious in the deleted anonymized dataset. [23] If this data is used to provide advertisements or to personalize product recommendations, re-identification can be largely harmless. What's dangerous is that data can and can sometimes be used to make inferences about future behavior or personal privacy, which can lead to the rejection of loans or jobs. According to Rocher (2019) [24], computer algorithms can identify 99.98% of Americans with only 15 attributes per person, without including names or other unique data. This shows how easy it is to re-identify using data.

Netflix Terms of Use

According to its privacy policy[25], Netflix gathers personal data such as names, email addresses, postal codes, payment information, phone numbers, device identifiers, geo-location, browser type, and other data that is automatically gathered when users use the service. Cookies and web beacons can be used to gather information about your interests when you use Netflix on your browser. Utilizing device identifiers, this remains true whether you use a streaming device, smartphone, or tablet. Netflix claims that demographic data like age and gender are not taken into account in the recommendations system when making decisions.[26]The Netflix terms further say that they share or disclose personal data with third parties (referred to as "Service Providers") to carry out tasks on their behalf or aid in the delivery of services. Additionally, the agreements say that Netflix does sell users' private information to third parties for financial gain.

Netflix uses personal data to distribute and customize marketing or advertising, as well as to track user responses to emails, marketing campaigns, and online advertisements on third-party websites. Additionally, according to Netflix's conditions, they take reasonable precautions to ensure the security and confidentiality of the personal data they acquire from Service users. According to Common Sense Media[27], an organization that reviews and provides ratings for media and technology with the goal of providing information on their suitability for children. They evaluated every each privacy rating. A higer rating (up to 100%) indicates that the product has more open privacy policies and better methods for safeguarding user data. The score is most used as a measure of the amount of further research someone will need to conduct in order to make an informed choice about a product. Netflix scored 50% and some concerns about privacy evaluation are as follows.

Responsible Use [28]

According to Zook (2017), "All big data research on social, medical, psychological, and economic phenomena engages with human subjects, and researchers have the ethical responsibility to minimize potential harm." [29] It is not known whether Netflix continuously monitors (viral) social interactions occurring in the real world and reflects them in uploaded works. In terms of reporting content, although the algorithm can distinguish, this does not filter or block content that is deemed inappropriate for the user.

Personal Data Gathering

In order to analyze personalized recommendations, users are frequently required to provide personal information, which puts them at risk for privacy violations [30]. Another issue that arises from data collection is that it does not immediately tell the user that your actions now will affect the Algorithm's decision. As human-AI interaction grows, Microsoft set 18 guidelines to help people keep in mind the implication of how AI should be designed. One of the ethical principles set by Microsoft emphasizes this in Initially: G1 Make clear what the system can do[31]. Netflix uses an AI system that influences based on user behavior. Although it's not unusual for discovery-based firms to use AI in this way, it does not explicitly let users know how their activities may affect others. Users can thumbs-up or thumbs-down media, for instance, but the system won't expressly tell you whether the action will affect what you watch in the future.[32]

Data Sharing: Protecting data from third parties

Collected data is shared with third parties such as your provider of streaming media devices, mobile phone carriers, television or internet service, and voice assistant platform providers, and Netflix indicates the purpose for sharing personal information with them.[33] However, it lacks adequate detail on which data is used for which purposes and how whether for research, product improvement, data analytics, and/or third-party marketing[34]. When it comes to third-party data access, they are even authorized to access a user's personal information.

Data Security: Protecting against unauthorized access

Users are required to create an account by authenticating additional personal information for the first time. However, this information is also unknown whether a third party is given access or protected by the company.[35] It is also unclear what jurisdiction a user's personal information is subject to.

Individual Control Issue

Netflix offers users only a limited set of controls over how algorithmic decision-making shapes their platform experience.[36] Taking the standards set by Microsoft's Guidelines for Human-AI Interaction into account, the ethics of Netflix's AI system deployment is questioned, and one of them concerns individual control: Over Time: G17 Provide Global Controls. [37] For an example, there is no way to turn off personalization on Netflix.[38] Will Richmond-Coggan, a technology and privacy specialist at Nottingham-based Freeths LLP, asserts, "There is very little scope to restrict the collection of device, connection, or activity information that is collected by default." People will have to decide if they are comfortable sharing that information with Netflix, but a lot of it is necessary for the platform's core functionality.

User Control to increase privacy

Despite this, Netflix allows users to file complaints through a grievance or remedy mechanism.[39] You can also disable your participation in test scenarios, such as advertisements for other Netflix shows, by going to Account > Settings > Test Participation.[40] In your profile under Marketing Communications, users can also check to make sure that you haven't chosen to let Netflix use your contact information "to send promotional communications on third party services".[41] In addition, the fact that the setting allows them to modify their information gives users some control over their personal information. As for copyrights, it is claimed to data or content collected from a user.

Unclear Ownership of the User Data

Ownership of the collected user data is not clearly revealed by Netflix. It is unclear whether Netflix would notify users of personal information shared with third parties when users request to have a look at it. It is not stated whether the company provides notification to the user when a government or legal request for the user's information is received. It has not been established whether users retain ownership of the user data that is collected and shared at this time and shared is not. In addition, it is unclear whether the scope or duration of copyright licenses for your data is limited.[42] Users of Netflix have minimal influence over how each feed's AI system is monitored and behaves.[43] Users can only choose whether the AI system arranges the "My List" category automatically or if they want to do it themselves.

References

  1. ↑ Netflix recommendation system: How it works. Reco AI. (2022, April 5). Retrieved February 8, 2023, from https://recoai.net/netflix-recommendation-system-how-it-works/#:~:text=What%20is%20the%20Netflix%20recommendation,based%20on%20users%20viewing%20preferences.
  2. ↑ Netflix recommendation system: How it works. Reco AI. (2022, April 5). Retrieved February 8, 2023, from https://recoai.net/netflix-recommendation-system-how-it-works/#:~:text=What%20is%20the%20Netflix%20recommendation,based%20on%20users%20viewing%20preferences.
  3. ↑ How Netflix's Recommendations System Works. Help Center. (n.d.). Retrieved January 25, 2023, from https://help.netflix.com/en/node/100639#:~:text=We%20estimate%20the%20likelihood%20that,preferences%20on%20our%20service%2C%20and.
  4. ↑ Netflix recommendation system: How it works. Reco AI. (2022, April 5). Retrieved February 8, 2023, from https://recoai.net/netflix-recommendation-system-how-it-works/#:~:text=What%20is%20the%20Netflix%20recommendation,based%20on%20users%20viewing%20preferences.
  5. ↑ YouTube. (2018). Netflix Research: Recommendations. YouTube. Retrieved February 8, 2023, from https://www.youtube.com/watch?v=f8OK1HBEgn0.
  6. ↑ Meltzer, R. (2020, July 7). How Netflix utilizes Data Science. Lighthouse Labs. Retrieved January 25, 2023, from https://www.lighthouselabs.ca/en/blog/how-netflix-uses-data-to-optimize-their-product
  7. ↑ Stoll, J. (2023, January 20). Netflix: Number of subscribers worldwide 2022. Statista. Retrieved February 8, 2023, from https://www.statista.com/statistics/250934/quarterly-number-of-netflix-streaming-subscribers-worldwide/
  8. ↑ Plummer, L. (2017, August 22). This is how Netflix's top-secret recommendation system works. WIRED UK. Retrieved January 25, 2023, from https://www.wired.co.uk/article/how-do-netflixs-algorithms-work-machine-learning-helps-to-predict-what-viewers-will-like.
  9. ↑ Eckhardt. (2018, November 13). Promise and peril for Machine Learning at Netflix. Technology and Operations Management. Retrieved February 10, 2023, from https://d3.harvard.edu/platform-rctom/submission/promise-and-peril-for-machine-learning-at-netflix/
  10. ↑ Eckhardt. (2018, November 13). Promise and peril for Machine Learning at Netflix. Technology and Operations Management. Retrieved February 10, 2023, from https://d3.harvard.edu/platform-rctom/submission/promise-and-peril-for-machine-learning-at-netflix/
  11. ↑ Gomez-Uribe, C. A., & Hunt, N. (2015). The Netflix Recommender System. ACM Transactions on Management Information Systems, 6(4), 1–19. https://doi.org/10.1145/2843948
  12. ↑ Netflix recommendation system: How it works. Reco AI. (2022, April 5). Retrieved February 8, 2023, from https://recoai.net/netflix-recommendation-system-how-it-works/#:~:text=What%20is%20the%20Netflix%20recommendation,based%20on%20users%20viewing%20preferences
  13. ↑ Netflix recommendation system: How it works. Reco AI. (2022, April 5). Retrieved February 8, 2023, from https://recoai.net/netflix-recommendation-system-how-it-works/#:~:text=What%20is%20the%20Netflix%20recommendation,based%20on%20users%20viewing%20preferences.
  14. ↑ Meltzer, R. (2020, July 7). How Netflix utilizes Data Science. Lighthouse Labs. Retrieved January 25, 2023, from https://www.lighthouselabs.ca/en/blog/how-netflix-uses-data-to-optimize-their-product
  15. ↑ How Netflix's Recommendations System Works. Help Center. (n.d.). Retrieved January 25, 2023, from https://help.netflix.com/en/node/100639#:~:text=We%20estimate%20the%20likelihood%20that,preferences%20on%20our%20service%2C%20and.
  16. ↑ Meltzer, R. (2020, July 7). How Netflix utilizes Data Science. Lighthouse Labs. Retrieved January 25, 2023, from https://www.lighthouselabs.ca/en/blog/how-netflix-uses-data-to-optimize-their-product.
  17. ↑ Staff, the P. N. O., Staff, D. P. I. P. and C. T. O., Smith, A., & Fair, L. (2022, June 13). Using artificial intelligence and algorithms. Federal Trade Commission. Retrieved January 25, 2023, from https://www.ftc.gov/business-guidance/blog/2020/04/using-artificial-intelligence-and-algorithms
  18. ↑ Staff, the P. N. O., Staff, D. P. I. P. and C. T. O., Smith, A., & Fair, L. (2022, June 13). Using artificial intelligence and algorithms. Federal Trade Commission. Retrieved January 25, 2023, from https://www.ftc.gov/business-guidance/blog/2020/04/using-artificial-intelligence-and-algorithms
  19. ↑ Staff, the P. N. O., Staff, D. P. I. P. and C. T. O., Smith, A., & Fair, L. (2022, June 13). Using artificial intelligence and algorithms. Federal Trade Commission. Retrieved January 25, 2023, from https://www.ftc.gov/business-guidance/blog/2020/04/using-artificial-intelligence-and-algorithms
  20. ↑ Staff, the P. N. O., Staff, D. P. I. P. and C. T. O., Smith, A., & Fair, L. (2022, June 13). Using artificial intelligence and algorithms. Federal Trade Commission. Retrieved January 25, 2023, from https://www.ftc.gov/business-guidance/blog/2020/04/using-artificial-intelligence-and-algorithms.
  21. ↑ IMDb.com. (n.d.). Ratings, reviews, and where to watch the best movies & TV shows. IMDb. Retrieved February 10, 2023, from https://www.imdb.com/
  22. ↑ Hosanagar, K., & Werbach, K. (2019, August 23). Is Data Privacy Real? Don't bet on it. Knowledge at Wharton. Retrieved February 10, 2023, from https://knowledge.wharton.upenn.edu/article/data-privacy-real-dont-bet/
  23. ↑ Hosanagar, K., & Werbach, K. (2019, August 23). Is Data Privacy Real? Don't bet on it. Knowledge at Wharton. Retrieved February 10, 2023, from https://knowledge.wharton.upenn.edu/article/data-privacy-real-dont-bet/
  24. ↑ Rocher, L., Hendrickx, J. M., & de Montjoye, Y.-A. (2019, July 23). Estimating the success of re-identifications in incomplete datasets using Generative Models. Nature News. Retrieved February 10, 2023, from https://www.nature.com/articles/s41467-019-10933-3/
  25. ↑ Netflix Help Center. (2022). Privacy statement. Privacy Statement. Retrieved February 10, 2023, from https://help.netflix.com/legal/privacy
  26. ↑ Netflix Help Center. (2022). Privacy statement. Privacy Statement. Retrieved February 10, 2023, from https://help.netflix.com/legal/privacy
  27. ↑ Common sense privacy evaluation for Netflix. The Common Sense Privacy Program. (n.d.). Retrieved January 26, 2023, from https://privacy.commonsense.org/evaluation/Netflix
  28. ↑ Common sense privacy evaluation for Netflix. The Common Sense Privacy Program. (n.d.). Retrieved January 26, 2023, from https://privacy.commonsense.org/evaluation/Netflix
  29. ↑ Zook, M., Barocas, S., boyd, danah, Crawford, K., Keller, E., Gangadharan, S. P., Goodman, A., Hollander, R., Koenig, B. A., Metcalf, J., Narayanan, A., Nelson, A., & Pasquale, F. (2017). Ten simple rules for responsible big data research. PLOS Computational Biology, 13(3). https://doi.org/10.1371/journal.pcbi.1005399
  30. ↑ DeLeon, H. (2019, April 24). The ethical and privacy issues of recommendation engines on media platforms. Medium. Retrieved January 25, 2023, from https://towardsdatascience.com/the-ethical-and-privacy-issues-of-recommendation-engines-on-media-platforms-9bea7bcb0abc
  31. ↑ Amershi, S., Weld, D., Vorvoreanu, M., Fourney, A., & Nushi, B. (n.d.). Guidelines for Human-AI Interaction. Retrieved 2023, from chrome-extension://efaidnbmnnnibpcajpcglclefindmkaj/https://www.microsoft.com/en-us/research/uploads/prod/2019/01/Guidelines-for-Human-AI-Interaction-camera-ready.pdf.
  32. ↑ Koo, M. (2019, September 15). Ethical Analysis of Netflix. Medium. Retrieved February 8, 2023, from https://medium.com/@koominji/ethical-analysis-of-netflix-7316183d09fa
  33. ↑ Common sense privacy evaluation for Netflix. The Common Sense Privacy Program. (n.d.). Retrieved January 26, 2023, from https://privacy.commonsense.org/evaluation/Netflix
  34. ↑ O'Flaherty, K. (2021, October 10). All the ways Netflix tracks you and what you watch. WIRED UK. Retrieved February 10, 2023, from https://www.wired.co.uk/article/netflix-data-tracking-privacy
  35. ↑ Common sense privacy evaluation for Netflix. The Common Sense Privacy Program. (n.d.). Retrieved January 26, 2023, from https://privacy.commonsense.org/evaluation/Netflix
  36. ↑ Why am I seeing this? New America. (n.d.). Retrieved February 10, 2023, from https://www.newamerica.org/oti/reports/why-am-i-seeing-this/case-study-netflix/
  37. ↑ Amershi, S., Weld, D., Vorvoreanu, M., Fourney, A., & Nushi, B. (n.d.). Guidelines for Human-AI Interaction. Retrieved 2023, from chrome-extension://efaidnbmnnnibpcajpcglclefindmkaj/https://www.microsoft.com/en-us/research/uploads/prod/2019/01/Guidelines-for-Human-AI-Interaction-camera-ready.pdf.
  38. ↑ O'Flaherty, K. (2021, October 10). All the ways Netflix tracks you and what you watch. WIRED UK. Retrieved February 10, 2023, from https://www.wired.co.uk/article/netflix-data-tracking-privacy
  39. ↑ Amershi, S., Weld, D., Vorvoreanu, M., Fourney, A., & Nushi, B. (n.d.). Guidelines for Human-AI Interaction. Retrieved 2023, from chrome-extension://efaidnbmnnnibpcajpcglclefindmkaj/https://www.microsoft.com/en-us/research/uploads/prod/2019/01/Guidelines-for-Human-AI-Interaction-camera-ready.pdf.
  40. ↑ O'Flaherty, K. (2021, October 10). All the ways netflix tracks you and what you watch. WIRED UK. Retrieved February 10, 2023, from https://www.wired.co.uk/article/netflix-data-tracking-privacy
  41. ↑ O'Flaherty, K. (2021, October 10). All the ways netflix tracks you and what you watch. WIRED UK. Retrieved February 10, 2023, from https://www.wired.co.uk/article/netflix-data-tracking-privacy
  42. ↑ Amershi, S., Weld, D., Vorvoreanu, M., Fourney, A., & Nushi, B. (n.d.). Guidelines for Human-AI Interaction. Retrieved 2023, from chrome-extension://efaidnbmnnnibpcajpcglclefindmkaj/https://www.microsoft.com/en-us/research/uploads/prod/2019/01/Guidelines-for-Human-AI-Interaction-camera-ready.pdf.
  43. ↑ Koo, M. (2019, September 15). Ethical Analysis of Netflix. Medium. Retrieved February 8, 2023, from https://medium.com/@koominji/ethical-analysis-of-netflix-7316183d09fa
Retrieved from "http://si410wiki.sites.uofmhosting.net/index.php?title=Netflix_Algorithm&oldid=116462"