Algorithmic logics of taste: Cultural taste and the Netflix recommender system Marie Fatima Iñigo Gaw A thesis submitted in fulfilment of the requirements for the degree of Master of Digital Communication and Culture Department of Media and Communications School of Literature, Art, and Media Faculty of Arts and Social Sciences The University of Sydney November 2019 Statement of originality This is to certify that to the best of my knowledge, the content of this thesis is my own work. This thesis has not been submitted for any degree or other purposes. I certify that the intellectual content of this thesis is the product of my own work and that all the assistance received in preparing this thesis and sources have been acknowledged. Marie Fatima Iñigo Gaw ii Acknowledgements Like algorithms, this research began as an ambiguous and messy project, buried under complex literature and convoluted thoughts. Through the help and support of mentors, peers and friends, I managed to weave everything together and craft something that I am genuinely proud of. Utmost gratitude to my research supervisor, Dr Justine Humphry, for the intellectual generosity and unwavering patience in helping me clarify, strengthen, sharpen my research. The time and support she gave me was beyond what was expected of her and was what inspired me to think and work even harder, especially during the difficult moments in the research process. I thank and appreciate the researchers at the STuF Lab led by Dr Marcus Carter for sharing new perspectives and constructive feedback on my research. The collective intelligence of the room never left me empty-handed. This research and my study at the university is supported by the Department of Foreign Affairs and Trade of the Government of Australia, through the program Australia Awards. I am indebted to the people of Australia for the opportunity to experience life abroad and attend one of the world’s most prestigious universities. To my friends—Nico Pablo, Tina Sablayan, Nathan Gatpandan, Reeneth Santos, Tristan Zinampan and Carmela Bangsal—for letting me bother them every time I need to sense-check my ideas and find the perfect words to capture those thoughts, thank you. iii A million thanks to my partner, Gabriel Murillo, for all the late-night consultations and relentless reminders to breathe when things get overwhelming. It is now my turn to cheer you on as you complete your master’s thesis. Lastly, to my family, Rosemarie Gaw, Charlotte Gaw and Nimfa Iñigo, who always let me pursue my passions even if it meant I would be away from them, I am grateful for all the love and support. iv Abstract Algorithms are new cultural intermediaries (Bourdieu, 1984) that shape contemporary cultural experiences and identities. Their obscurity and complexity, however, hinder us from understanding their logics and processes as tastemakers. This research investigates the algorithmic logics of taste of the Netflix recommender system (NRS) to theorise the NRS’s construction of taste and workings as a cultural intermediary. I adapt Taina Bucher’s (2016) technography as a methodological approach in studying algorithms beyond the ‘black box’, through the analysis of discursive materials and traces of algorithmic interactions with key social actors. Findings reveal that the NRS constructs taste as rules—universal, definite and durable assumptions about cultural identities and objects. They are enacted through the algorithmic infrastructure and constrain human agency through predefined choices without adequate mechanisms for negotiation. This post- hegemonic power (Lash, 2007) contributes to the reproduction of dominant social structures through new ways to interpellate and codify social categories that are the basis of cultural taste. Through their logics, algorithms as cultural intermediaries transcend being capital translators (Hutchinson, 2017) and encompass cultural production, distribution and consumption. Their limitations and fallibilities, however, open pathways for rejecting and subverting the algorithmic construction of taste. The study presents theoretical and empirical contributions to research on algorithmic cultures and cultural taste, as well as methodological innovation in studying socio-technical actors. I acknowledge that research on algorithms is v always partial and limited and thus, I prescribe directions for further research, with the intent to open a conversation on how algorithms can work better for/with humans. Keywords: algorithms, Netflix recommender system, cultural intermediaries, cultural taste, technography, discourse vi Contents Statement of originality ii Acknowledgements iii Abstract v Contents vii 1 Introduction 1 1.1 Research inquiry ................................................................................ 3 1.2 Significance and innovation .............................................................. 5 1.3 Considerations and limitations ......................................................... 6 2 Literature review 8 2.1 Algorithms as new cultural intermediaries ....................................... 8 2.2 Theorising cultural taste ................................................................... 11 2.3 The Netflix recommender system ................................................... 14 3 Methodology 18 3.1 Methodological approach: Technography ....................................... 18 3.2 Research methods ............................................................................ 19 3.2.1 Discourse analysis ................................................................. 19 3.2.2. Accretion measures ............................................................... 21 3.3 Data collection ................................................................................. 22 3.4 Data analysis .................................................................................... 24 3.5 Ethics statement .............................................................................. 26 vii 4 Findings 27 4.1 Discursive formations .................................................................... 27 4.1.1 Extraction: Implicit and explicit taste preferences ............. 28 4.1.2 Appraisal: Altgenres, taste communities, criteria and A/B testing ............................................................................ 29 4.1.3 Prediction: Recommendations, homepage personalisation and artwork and evidence personalisation ........................... 31 4.2 Discursive manifestations and contradictions ................................ 32 4.2.1 Issues and criticisms in media discourse ............................. 33 4.2.2 Netflix user interactions on Twitter ...................................... 35 5 Discussion and analysis 39 5.1 Construction of taste ........................................................................ 39 5.2 Algorithmic logics of taste ............................................................... 41 5.3 Algorithms as cultural intermediaries .............................................. 47 6 Conclusion and recommendations 51 References 56 Appendices 68 Appendix A: Discursive formations in the Netflix and media discourse 68 Appendix B: Issues and criticisms in the media discourse 75 Appendix C: Netflix user interaction coding scheme 81 Appendix D. Netflix user interaction geolocation on Twitter 88 viii Chapter 1 Introduction Since it launched its streaming business in 2007, Netflix has disrupted the way we access and consume television content. Core to its business is the Netflix recommender system (NRS), a set of algorithms that suggests content based on individuals’ taste preferences. Through the algorithms’ computational power, Netflix devised new mechanisms to refine its recommendations. Its vast library of content is organised into thousands of hyper-specific categories called altgenres (Madrigal, 2014). Rejecting demographic profiling, it created global taste communities as a way to distinguish its member base (Rodriguez, 2017). Recommendations have also become more granular as the algorithms customise title sequencing, artwork, search, and so on. Algorithmic recommendations are a compelling part of the Netflix experience, but they are not free of criticisms. Controversies of alleged racial and gender-based recommendations surrounded Netflix last year (Berkowitz, 2018; Ha, 2019). Conflicts between Netflix and Hollywood emerged with the purported ascendancy of algorithmic rationality over creative decisions (Ramachandran & Flint, 2018). Critics also contend that recommendations have not been responsive to users’ taste preferences (Diaz, 2018). In all these controversies, Netflix executives upheld the algorithms’ neutrality by reiterating that the recommendations are exclusively based on users’ taste preferences and viewing habits. The NRS represents the distinct ways algorithmic machines are shaping cultural encounters and facilitating everyday cultural processes. Algorithms are a new kind 1 of cultural intermediary (Morris, 2015; Gillespie, 2016; Hutchinson, 2017), which Bourdieu (1984) defines as entities that create and manipulate meanings attached to commodities to facilitate their movement in the market. Negus (2002) builds on this definition by positioning cultural intermediaries as the bridge between the production and consumption of symbolic goods. Hutchinson (2017) designates them as capital translators, transforming capital from one form to another to create new value. However, algorithms are inherently complex and are concealed in proprietary ‘black boxes’ (Bucher,