{"id":5115,"date":"2019-03-20T18:42:04","date_gmt":"2019-03-20T22:42:04","guid":{"rendered":"http:\/\/studentwork.prattsi.org\/foundations\/?p=5115"},"modified":"2019-03-20T18:42:11","modified_gmt":"2019-03-20T22:42:11","slug":"how-netflix-learns-what-you-like","status":"publish","type":"post","link":"https:\/\/studentwork.prattsi.org\/foundations\/2019\/03\/20\/how-netflix-learns-what-you-like\/","title":{"rendered":"How Netflix Learns What You Like"},"content":{"rendered":"\n

On Thursday, February 28th<\/sup>, NYU Tandon School of Engineering held a live streaming event featuring a talk given by Netflix\u2019s Director of Machine Learning, Tony Jebara<\/a>. The topic covered was \u201cMachine Learning for Personalization\u201d, which Jebara provided company use cases and solutions for content personalization.<\/p>\n\n\n\n

\"\"
Netflix Director of Machine Learning, Tony Jebara
<\/figcaption><\/figure><\/div>\n\n\n\n

Netflix<\/a>,\na streaming media-service, is well regarded within the machine learning field\nfor developing impressive machine learning models that incorporate advanced\nfeedback mechanisms to train and improve those models. <\/p>\n\n\n\n

According to the Director of Machine Learning, Tony Jebara, every Netflix user\u2019s experience is unique across a range of personalized content. A few examples of personalized content provided by Jebara were rankings, homepage generation, promotions, image selections, searches, advertisement displays, and push notifications. <\/p>\n\n\n\n

Content Personalization<\/em><\/h3>\n\n\n\n

Content\npersonalization is a technique leveraged by many companies, across many\nindustries, for the business of either creating content, distributing it or\nboth. Content encompasses everything from online articles to advertisements. In\nDigital Disconnect<\/em>, McChesney\ndescribes that the popular digital method \u201cpersonalizes content for\nindividuals, and the content is selected based on what is considered most\nlikely to assist the sale\u201d (p.157).<\/p>\n\n\n\n

Entrepreneur<\/em><\/a> lauded Netflix and other media companies who are successfully leveraging machine learning to develop custom experiences but notes a dichotomy which plagues user\u2019s and their preferences. The trade-off between conveniently custom experiences or inconveniently anonymous reintroductions. On one side, users face issues surrounding privacy or unpleasant information dictation.<\/p>\n\n\n\n

Opposite to their praises as personalization gurus, Fast Company<\/em><\/a> <\/em>highlighted some of the negative criticisms Netflix has also received. When companies curate the content users consume, there’s a risk of receiving biased information whether it be political or racial. Berkowitz opens with, \u201cHow companies advertise to you says a lot about how they see you\u201d when referring to the racial bias in the algorithms used by not only Netflix, in this case, but many of the other companies working to deploy advanced content personalization algorithms.<\/p>\n\n\n\n

“Filter Bubbles”<\/em><\/h3>\n\n\n\n

Regarding\nthe politically charged dictation of content, Castells remarks, \u201cThe networks\nthemselves reflect and create distinctive cultures. Both they and the traffic\nthey carry are largely outside national regulation. Our dependence on the new\nmodes of informational flow gives to those in a position to control them\nenormous power to control us. The main political arena is now the media, and\nthe media are not politically answerable\u201d (p. 34).<\/p>\n\n\n\n

McChesney adds how these practices also lead to an issue he considers the \u201cpersonalization bubble\u201d or what he specifically alludes to as the \u201cfilter bubble\u201d (p.157). Users are trapped in an experience they believe to be unique or new but is perpetuated by the same content delivery\u2014just done differently (p. 70).<\/p>\n\n\n\n

McChesney references Eli Pariser\u2019s The Filter Bubble: How the New Web is Changing What We Read and How We Think<\/em> when stating, \u201cPariser’s Filter Bubble documented how the Internet is quickly becoming a personalized experience wherein people get different results on Google searches for identical queries, based on their history\u201d (p.157).<\/p>\n\n\n\n

When Netflix Intervenes<\/em><\/h3>\n\n\n\n

In\nhis talk, Jebara claimed that \u201cprediction is valuable but actual intervention\nis what we want to understand.\u201d Their algorithms are two-fold\u2014ensuring that\nexperiences are uniquely specific without providing recommendations that are\ntoo specific, which may lead to either a negative user experience and a\npotential unsubscribe from the service.<\/p>\n\n\n\n

\"\"<\/figure><\/div>\n\n\n\n

We\u2019ve all experienced moments of interacting with a digital platform that, over time and with enough data aggregation, begins to recommend content or display ads across devices and sites outside the ownership of the originating platform. If frightened enough, we may have even gone as far as to deleting our browser cookies, adjusting our privacy settings or even unsubscribing from the service.<\/p>\n\n\n\n

Algorithm Feedback<\/em><\/h3>\n\n\n\n

Jebara\nmentioned that a multitude of mixed-method machine learning algorithms are\nimplemented to hone everything from predictive analytics and image curation to user-enforced\nrestrictions and feedback mechanisms.<\/p>\n\n\n\n

Jebara\ndescribed their method take rate<\/em> as a\ncuratorial feedback strategy which tests different personalization experiences\non several users to determine which of the content shown resulted in an actual\nviewing.<\/p>\n\n\n\n

\"\"<\/figure><\/div>\n\n\n\n
\"\"<\/figure><\/div>\n\n\n\n

This\nstrategy uniquely prefers the measurement of the number of viewers that\nstrategy worked for over the number of viewers a specific piece of content was\nshown to. Jebara noted this method enables Netflix experts to learn from users\nby letting them show<\/em> what content\nthey prefer and in which ways they\u2019re drawn to recommendations. <\/p>\n\n\n\n

User Generated<\/em> Feedback<\/em><\/h3>\n\n\n\n

This is a major shift from their previous user experience of providing users with the ability to ranking rank content using a star ranking system. Overtime and through observation, Netflix realized they couldn\u2019t rely on that ranking system as a source of truth for which content users ranked highly versus which they\u2019d prefer to watch. Jebara added users were not truthful in their telling<\/em> of which content they preferred. Shifting away from user interaction to user observation has enabled a greater foundation for developing recommendation systems. <\/p>\n\n\n\n

Conclusion<\/em><\/h3>\n\n\n\n

As content personalization algorithms advance, consumers will become a more passive actor in teaching content personalization algorithms. Every attempt at restricting interaction with such algorithms will lead only to yet another loophole identified by machine learning experts. How those companies manage those algorithms and exploit those loopholes are examples of the digital power dynamic which exists between the content generators and the content consumers. <\/p>\n\n\n\n

References<\/u>:<\/p>\n\n\n\n

Berkowitz, Joe. \u201cIs Netflix racially personalizing\nartwork for its titles?One writer\u2019s experience with\nNetflix\u2019s title art has us wondering whether the company is quietly using race\nin its algorithm for visually recommending films<\/em>\u201d. Fast Company (2018). https:\/\/www.fastcompany.com\/90253578\/is-netflix-racially-personalizing-artwork-for-its-titles<\/a><\/p>\n\n\n\n

Castells,\nManuel. The Information Age: Economy,\nSociety, and Culture<\/em>. Wiley-Blackwell (2010). https:\/\/lms.pratt.edu\/pluginfile.php\/876462\/mod_resource\/content\/1\/manuel_castells_the_rise_of_the_network_societybookfi-org.pdf<\/a><\/p>\n\n\n\n

Chmielewski,\nDawn C. \u201cNetflix\u2019s Use of Artwork\nPersonalization Attracts Online Criticism\u201d. Deadline (2018).  https:\/\/deadline.com\/2018\/10\/netflixs-artwork-personalization-attracts-online-criticism-1202487598\/<\/a><\/p>\n\n\n\n

McChesney,\nRobert W. Digital Disconnect<\/em>. The New\nPress (2013): 63-171. <\/p>\n\n\n\n

Wirth,\nKarl. \u201cNetflix Has Adopted Machine\nLearning to Personalize Its Marketing Game at Scale: Here’s how you can\nhumanize marketing strategies<\/em>\u201d.<\/em>\nEntrepreneur (2018). https:\/\/www.entrepreneur.com\/article\/311931<\/a><\/p>\n","protected":false},"excerpt":{"rendered":"

On Thursday, February 28th, NYU Tandon School of Engineering held a live streaming event featuring a talk given by Netflix\u2019s Director of Machine Learning, Tony Jebara. The topic covered was \u201cMachine Learning for Personalization\u201d, which Jebara provided company use cases and solutions for content personalization. Netflix, a streaming media-service, is well regarded within the machine […]<\/p>\n","protected":false},"author":610,"featured_media":5150,"comment_status":"open","ping_status":"open","sticky":false,"template":"","format":"standard","meta":{"footnotes":""},"categories":[2,3,245],"tags":[372,373],"class_list":["post-5115","post","type-post","status-publish","format-standard","has-post-thumbnail","hentry","category-articles","category-event-reviews","category-sula","tag-content-personalization","tag-machine-learning"],"jetpack_featured_media_url":"https:\/\/studentwork.prattsi.org\/foundations\/wp-content\/uploads\/sites\/5\/2019\/03\/IMG_2694-1024x576-1.png","_links":{"self":[{"href":"https:\/\/studentwork.prattsi.org\/foundations\/wp-json\/wp\/v2\/posts\/5115","targetHints":{"allow":["GET"]}}],"collection":[{"href":"https:\/\/studentwork.prattsi.org\/foundations\/wp-json\/wp\/v2\/posts"}],"about":[{"href":"https:\/\/studentwork.prattsi.org\/foundations\/wp-json\/wp\/v2\/types\/post"}],"author":[{"embeddable":true,"href":"https:\/\/studentwork.prattsi.org\/foundations\/wp-json\/wp\/v2\/users\/610"}],"replies":[{"embeddable":true,"href":"https:\/\/studentwork.prattsi.org\/foundations\/wp-json\/wp\/v2\/comments?post=5115"}],"version-history":[{"count":3,"href":"https:\/\/studentwork.prattsi.org\/foundations\/wp-json\/wp\/v2\/posts\/5115\/revisions"}],"predecessor-version":[{"id":5149,"href":"https:\/\/studentwork.prattsi.org\/foundations\/wp-json\/wp\/v2\/posts\/5115\/revisions\/5149"}],"wp:featuredmedia":[{"embeddable":true,"href":"https:\/\/studentwork.prattsi.org\/foundations\/wp-json\/wp\/v2\/media\/5150"}],"wp:attachment":[{"href":"https:\/\/studentwork.prattsi.org\/foundations\/wp-json\/wp\/v2\/media?parent=5115"}],"wp:term":[{"taxonomy":"category","embeddable":true,"href":"https:\/\/studentwork.prattsi.org\/foundations\/wp-json\/wp\/v2\/categories?post=5115"},{"taxonomy":"post_tag","embeddable":true,"href":"https:\/\/studentwork.prattsi.org\/foundations\/wp-json\/wp\/v2\/tags?post=5115"}],"curies":[{"name":"wp","href":"https:\/\/api.w.org\/{rel}","templated":true}]}}