Behaviour & Information Technology

ISSN: (Print) (Online) Journal homepage: https://www.tandfonline.com/loi/tbit20

How using various platforms shapes awareness of

algorithms

Johan Espinoza-Rojas, Ignacio Siles & Thomas Castelain

To cite this article: Johan Espinoza-Rojas, Ignacio Siles & Thomas Castelain (2022): How using
various platforms shapes awareness of algorithms, Behaviour & Information Technology, DOI:
10.1080/0144929X.2022.2078224

To link to this article: https://doi.org/10.1080/0144929X.2022.2078224

Published online: 17 May 2022.

Submit your article to this journal

View related articles

View Crossmark data

Full Terms & Conditions of access and use can be found at

https://www.tandfonline.com/action/journalInformation?journalCode=tbit20
BEHAVIOUR & INFORMATION TECHNOLOGY
https://doi.org/10.1080/0144929X.2022.2078224

RESEARCH ARTICLE

How using various platforms shapes awareness of algorithms

a b c
Johan Espinoza-Rojas , Ignacio Siles and Thomas Castelain
a
Master in Cognitive Sciences, University of Costa Rica, San José, Costa Rica ; bSchool of Communication, University of Costa Rica, San José,
Costa Rica; cDepartment of Psychology, Institute of Educational Research, University of Girona, Girona, Spain

ABSTRACT ARTICLE HISTORY

This paper examines how the use of multiple platforms is tied to awareness of algorithms. It builds Received 20 November 2021
on the premise that users interact with ecologies or environments of technologies rather than Accepted 10 May 2022
single platforms. The study also supplements work on algorithmic awareness by implementing
KEYWORDS
a mixed-method study to account for how Costa Rican users of Netflix and Spotify understood Algorithms; awareness;
and related to the algorithms of these platforms. This study combined a survey of 258 cognition; Latin America;
participants and 21 in-depth semistructured interviews. Findings demonstrate that multi- Netflix; Spotify
platform users were more aware of algorithms and carried out more practical actions to obtain
algorithmic recommendations than single-platform users. Although user type did not predict
participants’ attitudes towards algorithmic recommendations, higher levels of awareness were
associated with more positive attitudes towards algorithms. The study also shows that
differences in levels of awareness explained users’ emotional arousal derived from algorithms.

1. Introduction
relationship with media technologies should be under-
Algorithms are surrounded by secrecy. Scholars usually stood as part of larger environments where life unfolds
employ terms such as ‘black boxes’ or ‘opacity’ to (Boczkowski and Mitchelstein 2021). Although these
describe their operation and the social consequences of ideas have been established in the literature for some
their uptake (Burrell 2016; Pasquale 2015). This opacity time, most studies of algorithmic awareness to date
makes algorithms difficult to understand and regulate have tended to focus primarily on how people interpret
(Christin 2020). Accordingly, most research suggests and understand the algorithmic operations of one indi-
that users are largely unaware of algorithms (Eslami vidual technology at a time, be it social media platforms,
et al. 2015; Powers 2017; Schwartz and Mahnke 2021). news apps, or streaming devices. Recent research has
Against this background, a growing body of work has emphasised the need to investigate more than one plat-
sought to make sense of how users understand, think form since ‘algorithmic [operations] frequently [occur]
of, relate to, and feel about algorithms (Ytre-Arne and across platforms and contexts’ (Büchi et al. 2021, 13).
Moe 2021). The premise behind such studies is that, Building on this premise, we show that the use of mul-
even if algorithms are ‘opaque,’ they are not necessarily tiple platforms is tied to more awareness of algorithms,
‘unknowable’ (Bucher 2018). In other words, users more practical actions to receive recommendations,
understand and relate to them in specific ways. more positive attitude to platforms, and more emotional
This article contributes to this body of work by exam- arousal derived from their use.
ining how the use of multiple platforms shapes how Definitions of user awareness vary. For the purposes
people understand, relate to, and feel about the algor- of this study, we draw on Dourish and Bellotti’s (1992)
ithms of Netflix and Spotify. People typically don’t use notion that awareness is ‘an understanding of the activi-
only one platform but rather multiple interrelated arti- ties of others, which provides a context for your own
facts (Boczkowski and Mitchelstein 2021). Madianou activity. This context is used to ensure that individual
and Miller’s (2013) notion of ‘polymedia’ emphasises contributions are relevant to [an] activity as a whole,
how ‘people use a constellation of different media as and to evaluate individual actions with respect to […]
an integrated environment in which each medium goals and progress’ (107). We argue that the use of mul-
finds its niche in relation to the others’ (3). In a similar tiple platforms shapes users’ awareness and attitudes
manner, scholars have pointed out that people’s towards algorithms.

CONTACT Thomas Castelain [email protected] Department of Psychology, Institute of Educational Research, University of Girona, Girona,
Spain
Supplemental data for this article can be accessed online at https://doi.org/10.1080/0144929X.2022.2078224.
© 2022 Informa UK Limited, trading as Taylor & Francis Group
2 J. ESPINOZA-ROJAS ET AL.

We draw on a mixed-method study to account for how Bucher (2017) theorised imaginaries as ‘ways of thinking
people understood Netflix’s and Spotify’s algorithms. about what algorithms are, what they should be, how
Most previous studies of people’s relationship with algor- they function, and what these imaginations, in turn,
ithms have adopted either quantitative approaches (to make possible’ (113). In a similar manner, ‘folk theories’
augment generalizability capacity) or qualitative methods refer to the intuitive theories that people hold to account
(to add nuance and depth to findings). As a supplement, for how algorithms work (DeVito, Gergle, and Birnholtz
we combined a survey of 258 participants and 21 in-depth 2017; Ytre-Arne and Moe 2021). Together, these concepts
semistructured interviews. Survey data allowed us to point to how users develop their own explanations of what
examine the differences between multi-platform and indi- algorithms are and how they operate. These intuitive
vidual-platform users. Interviews then offered an oppor- explanations are critical in that they shape how users incor-
tunity to further understand how users made sense of porate algorithms in their daily lives. In other words, ima-
their interactions with algorithmic platforms. ginaries and folk theories speak to the issue of agency: they
The study was conducted in Costa Rica. Several fac- partly explain why users behave in certain ways.
tors justify this case selection. Costa Rica has all the Research has distinguished between various forms of
characteristics that have made Latin America an attrac- algorithmic awareness. Koenig (2020) defined ‘basic
tive place for algorithmic platforms such as Netflix and awareness’ as ‘a basic understanding of what an algor-
Spotify: it has a relatively large middle-class, reliable tel- ithm does (as a computational process)’ (5). For Koen-
ecommunications infrastructure, and high Internet ing, ‘critical awareness’ refers instead to explanations of
access rates (Red 506 2018). Latinobarómetro (2018) why algorithms function the way they do and the impli-
data shows that the country leads the use of numerous cations of their operation. Finally, Koenig posited the
social media platforms in the region. The country also term ‘rhetorical awareness’ to consider how users of
ranks high in connectivity rates in Latin America. Con- platforms understand their relationship with algorithms
sidering Costa Rica’s case thus allows us to discuss as a mutually influencing process. Thus, for Koenig
findings from one of the region’s most digital countries (2020, 6), ‘A fluidity occurs in the connection between
in terms of infrastructure, culture, and use. user and algorithmic machine, and this connection
We focused on Netflix and Spotify because of their causes action. […] Rhetoric, in this sense, is more
importance in the consumption of cultural content in about the interconnected components of the rhetorical
Latin America and the centrality of algorithms in their act.’ In a similar manner, Siles, Espinoza-Rojas and col-
operation (Eriksson et al. 2019; Lobato 2019). Both plat- leagues (2019) theorised this rhetorical relationship as a
forms have been available in the region since the early ‘mutual domestication’ process: ‘users incorporate algo-
2010s decade, which allowed us to consider platforms rithmic recommendations into everyday life as much as
with relatively long trajectories and an important num- the platform works to colonise users and turn them into
ber of users in the country (Red 506 2018). By consider- ideal consumers through its algorithms’ (500).
ing platforms that offer relatively distinct types of Users’ awareness comes from a variety of sources.
products (movies, series, music, and podcasts) we Research has sought to elucidate the diversity of criteria
sought to account for diversity while retaining the and factors involved in how users become aware of
focus on the domain of entertainment. algorithms. Most studies suggest that awareness comes
from practical action: the more people use specific fea-
tures of platforms, the more aware they become of
2. Becoming aware of algorithms
how algorithms operate. In short, awareness is the result
Studies on users’ awareness of algorithms have focused of an ‘active engagement’ with platforms (Eslami et al.
on various issues that we discuss next: how users make 2015). DeVito and colleagues (2018) also demonstrated
sense of algorithms in their daily interactions with plat- that users’ understanding of algorithms is shaped by fac-
forms, what sources are involved in the development of tors that transcend technology itself. In short, these
their awareness, and how users develop affective atti- authors write, ‘people consistently tend to use knowl-
tudes towards algorithms. edge in the world […] to supplement existing knowl-
edge of [an] artifact or system’ (DeVito et al. 2018, 3).
Referring specifically to the case of folk theories, they
2.1. Awareness and active engagement
theorise these factors as ‘endogenous’ and ‘exogenous’
First, scholars have analyzed how users make sense of to platforms. Multi-platform use sits at the intersection
algorithms. Studies have demonstrated that users develop of endogenous and exogenous factors in that it allows
a repertoire of ‘algorithmic imaginaries’ and ‘folk theories’ mobilising awareness generated through engagement
in their relationship with platforms (Siles et al. 2020). with various algorithmic devices.
 BEHAVIOUR & INFORMATION TECHNOLOGY 3

Considering these reflections, our first hypotheses provides gratifications to users’ needs to better under-
state: stand their relationship with algorithms and platforms.
To this end, we examined three types of needs that can
(H1) Multi-platform users will be more aware of algor-
ithms than single-platform users. explain the relationship between the type of user and the
awareness of algorithms: affective, cognitive, and
(H2) Multi-platform users will carry out more practical integrative.
actions to obtain algorithmic recommendations than Affective needs deal with issues of emotion, attitude,
single-platform users.
and feeling (Tefertiller 2018). Accordingly, it is
expressed when users turn to algorithmic platforms
2.2. Attitudes and emotional arousal for entertainment, relaxation, and leisure. Cognitive
needs refer to the acquisition of knowledge and infor-
Scholars have also analyzed the ties between users’
mation (Hussain, Shabir, and Taimoor Ul 2020). In
awareness and their attitudes towards algorithms
the case of our research, this need is expressed by the
(Eslami et al. 2019; Gran, Booth, and Bucher 2021;
tendency to search for information that allows users
Grzymek and Puntschuh 2019). As Bucher (2017) con-
to understand how exactly algorithms work. Finally,
tends, users’ relationship with algorithms is profoundly
integrative needs centre on people’s self-understanding
affective. Bucher’s notion of the imaginary stresses the
as part of social groups (Sjöblom 2015). People’s use
significance of people’s feelings and emotions when
of algorithmic platforms to sustain social relationships
they use certain algorithmic platforms and how this
and perform certain identities is an expression of this
matters for their actions (Bucher 2017). In a similar
type of need in the case of algorithms.
manner, Siles and colleagues (2021, 2023) drew on
To test our four hypotheses, we relied on a mixed
actor-network theory to theorise the user-algorithm
methodology. First, we conducted an online survey to
relationship as an affective attachment. Considering
determine how types of users, gratifications sought,
the case of TikTok, they argued that awareness of algor-
age, and gender of participants shaped their experience
ithms shaped users’ attachment to this app. Awareness
with algorithms. Complementarily, we carried out in-
provided users with a context for understanding their
depth interviews to have a finer-grained description of
capacities in relation to algorithms. As a result, users
the phenomena under study.
enacted various practical roles to maintain the affect
associated with personalised content on the TikTok.
The link between awareness and users’ attitudes 3. Methods
towards algorithms can also be discussed as an instance
of emotional arousal. This notion refers to the emotions 3.1. Ethical statement and data availability
that platforms produce or stimulate on users. Studies This research was approved by the Institutional Review
have argued that content with strong emotional arousal Board of the University of Costa Rica. All participants
tends to lead to more awareness in users (Anderson gave their written online informed consent. All the
2005; Sheth and Pham 2008). In the case of algorithmic materials, the database, the R script and the detail of
platforms, some preliminary work has shown that users the analysis are available in Open Science Framework
favour recommendations that take into account their (OSF): https://doi.org/10.17605/OSF.IO/R9TAV.
emotional context rather than purely sociodemographic
variables (Deng et al. 2015).
Building on the insights generated by this body of 3.2. Survey
work, we propose the following hypotheses:
3.2.1. Participants
(H3) Multi-platform users will have more positive atti- Participants were users of Netflix and Spotify and were
tudes towards algorithmic recommendations than categorised as single-platform users (regardless of which
single-platform users. one) and multi-platform users (both platforms). They
(H4) Multi-platform users will have higher levels of were recruited through a call for participation shared
emotional arousal from algorithms than single-plat- on institutional social media profiles and snowball
form users. sampling techniques. To avoid possible bias toward
technical knowledge regarding algorithms, participants
with a computer science background or related fields
2.3. Gratifications and needs
were excluded from the study. The final sample was
Along with each of the four hypotheses, it was con- constituted by 258 participants (167 of whom identified
sidered important to think about how digital media as women, M = 29.6 years-old, SD = 9.08, range [18:65]),
4 J. ESPINOZA-ROJAS ET AL.

166 multi-platform users and 92 single-platform users. to indicate and enumerate what kind of practices
Most of these participants (98%) resided in the Central characterised their usage of these platforms.
Valley region of Costa Rica, an area with higher popu- To examine emotional arousal, we designed a scale
lation density and high Internet access (Instituto Nacio- with four items for each platform based on the work
nal de Estadística y Censo 2020). All the participants conducted by Siles and colleagues (2019). Participants
had formal education and most had a university degree were invited to select the option that best represented
(63%). More than half of the subjects (61%) reported their opinions (from 1, ‘I do not identify at all’ to 5, ‘I
they had been users of these platforms for more than completely identify’) regarding items related to their
three years. usage practices and actions tied to their emotions and
needs while interacting with these platforms. We
asked about their tendency to select content based on
3.2.2. Material their moods, to repeat certain content to relive an
We designed an instrument to assess participants’ emotion, and their opinions regarding the accuracy of
awareness of algorithms, the number of actions they algorithms to recommend them content that matched
deploy to receive specific recommendations, their atti- their affective states. Higher scores indicate a higher
tudes towards algorithmic recommendations, the level of emotional arousal (α = 0.7 for Netflix, α = 0.7
emotional arousal that resulted from these recommen- for Spotify).
dations and the integrative, affective, and cognitive Finally, to measure the gratifications Netflix and
needs that platforms provided to users (see Supplemen- Spotify provide to users, we adapted the scale developed
tary Materials). by Tefertiller (2017). We built three scales to assess par-
To measure awareness, we developed a scale consist- ticipants’ integrative, affective, and cognitive needs. The
ing of six items based on a preliminary study (Siles et al. integrative needs scale was designed to understand how
2019). Participants were invited to select the option that needs for social belonging shaped participants’ use of
best represented their opinions (from 1, ‘I do not ident- platforms (α = 0.79 for Netflix, α = 0.72 for Spotify).
ify at all’ to 5, ‘I completely identify’) regarding algorith- The affective needs scale sought to capture how users
mic operations and their practices with these streaming were entertained by platforms (α = 0.74 for Netflix, α
services. Questions centred on their awareness about = 0.69 for Spotify). And the cognitive needs scale
changes in content recommendations on these plat- focused on users’ satisfaction in learning and experien-
forms, the extent to which they preferred to receive cing new things (α = 0.86 for Netflix, α = 0.86 for Spot-
algorithmic recommendations or not, their knowledge ify). Participants were asked to identify themselves
about personalisation mechanisms in these platforms, (from 1, ‘I do not identify at all’ to 5, ‘I completely ident-
their actions to receive or avoid specific algorithmic rec- ify’) to three items for each scale.
ommendations, and their opinions about what they per-
ceived to be failed recommendations that did not match
3.2.3. Procedure
their preferences. Higher scores indicate a higher level
A survey in Spanish was self-administered online. One
of awareness (α = 0.58 for Netflix, α = 0.7 for Spotify).
part of the survey was related to Netflix and the other
One item of each scale (Item 2 for Netflix, and Item 5
part was devoted to Spotify. Single-platform users
for Spotify) was dropped due to a low item-test corre-
were invited to answer the part that corresponded to
lation (r < 0.20).
the specific platform they used, whereas multi-platform
To measure the attitudes toward recommendations,
users completed both parts. The order in which the sec-
we adapted the Likert scale developed by Thurman
tions of the survey were presented to multi-platform
et al. (2019). Participants were invited to assess their
users was varied between participants.
attitudes toward Netflix or Spotify’s algorithmic rec-
ommendations by identifying themselves (from 1, ‘I
do not identify at all’ to 5, ‘I completely identify’) 3.2.4. Data treatment for quantitative analyses
based on their trust of algorithmic recommendations, To test H1, H3, and H4 we calculated a score for each
the use of the data collected by platforms, and their participant. This score corresponds to the mean of the
expectations regarding algorithmic personalisation. answers for each scale. For multi-platform users, we
Higher scores indicate a more positive attitude towards computed the mean of the two scales. To test H2, we
algorithmic recommendations (α = 0.8 for Netflix, α = first enumerated each action for each participant and
0.84 for Spotify). then created a score that represented the sum of these
To assess the actions undertaken by users to receive actions. We also calculated the mean of the two scales
specific algorithmic recommendations, we asked them for multi-platform users.
 BEHAVIOUR & INFORMATION TECHNOLOGY 5

3.3. Interviews Our strategy combined individual and collective

rounds of coding. A first round of coding was conducted
3.3.1. Participants
individually to understand how differences between
The participants recruited for the interviews were indi-
single-platform and multi-platform users materialised
viduals who took part in the survey. Among the 117 par-
in discourse and in practice. Accordingly, in this phase
ticipants who agreed to be interviewed, we selected 21
we focused on identifying how each group of users
for an in-depth, semi-structured conversation (10
showed awareness of algorithms, the type of actions
people identified as women, M = 29.5 years-old, SD =
they carried out to obtain algorithmic recommendations,
11.2, range [19:65]). Eleven were multi-platform users
and what kinds of attitudes and emotions characterised
and 10 were single-platform users. Most of the partici-
their relationships with these platforms.
pants (86%) lived in the Central Valley region of
Second, we worked collectively to refine the different
Costa Rica but had different educational backgrounds.
categories developed in the first round of coding.
Among interviewees, 5 had no formal education or
During this second phase, we established relationships
degree, 6 had some form of college education, 8 had a
between the main patterns identified during the initial
university degree, and 2 had graduate-level education.
round of coding. This allowed us to better understand
how forms of awareness, user actions, attitudes, and
3.3.2. Material emotions were tied to each other.
We developed an interview protocol to obtain more Finally, during the third round of coding, we collec-
nuanced and detailed information about our main vari- tively aggregated the data into broader categories that
ables of interest. We focused on better understanding captured the main patterns. This phase was devoted to
participants’ awareness of algorithms. Most questions building the categories of ‘single-platform user’ and
centred on the actions taken by individuals to deal ‘multi-platform user,’ to comparing them along the
with algorithmic recommendations; type of content dimensions proposed in our hypotheses, and to incor-
that platforms suggested to them; the kind of rec- porating the role of context (in this case Costa Rican
ommendations that best matched their preferences; society) in the construction of these categories, as
and their experiences with recommendations that they advised by grounded theory proponents.
did not like.
4. Results
3.3.3. Procedure
To take advantage of the complementarity of both
Due to restrictions associated with the COVID-19 pan-
methods we employed, data collected through the sur-
demic, interviews were conducted via Zoom and lasted
vey and in-depth interviews will be combined to discuss
an average of 40 min. Interviews were conducted in
each hypothesis of our study.
Spanish (all translations are our own). After answering
some initial questions, participants were invited to
open their Netflix/Spotify account, describe the content 4.1. Awareness
available; and explain why they thought they had
To test our first hypothesis, we conducted a multiple lin-
received specific recommendations. We implemented
ear regression analysis with user type (single-platform
a ‘think aloud protocol’ to discuss the configuration of
users coded as 0 and multi-platform users coded as 1),
each user’s ‘profile’ and invited interviewees to elaborate
affective needs, cognitive needs, integrative needs, gen-
on these configurations.
der (male coded as 0 and female coded as 1), and age
as predictors of participants’ awareness (see SM, Table
3.3.4. Data treatment for qualitative analyses 1 and Figure 1). The overall regression was statistically
We drew on grounded theory analytical procedures to significant (R 2 = 0.14, F(6, 251) = 8.24, p < .001) and
analyze in-depth interviews. We implemented revealed that multi-platform users were more aware
grounded theory primarily as a set of coding techniques than single-platform users (β = −0.26, p < .01), that
to identify patterns in the data and to develop in a sys- higher scores in cognitive needs were associated with
tematic way the categories we had recognised through higher levels of awareness (β = 0.16, p < .01), that male
the survey. This is consistent with research designs participants showed higher scores in Awareness (β =
that aim to develop ‘a close-up illustration of a bigger 0.29, p < .01) and that an increase in age was associated
picture’ (Mason 2006a) by combining methods in with a decrease in awareness (β = −0.01, p < .05).
ways that allow for more ‘richness, depth, complexity, Data from interviews help to better understand these
and nuance’ in explanations (Mason 2006b, 22). findings. Both single-platform and multi-platform users
6 J. ESPINOZA-ROJAS ET AL.

tended to singularise algorithms by referring to them as The use of one platform often operated as what partici-
a unitary, monolithic entity: ‘the algorithm.’ Single-plat- pants referred to as the ‘initial algorithmic experience,’
form users employed the term ‘algorithm’ mostly to which was then extrapolated to other platforms that
explain how they thought recommendations and the users thought worked similarly. During the interview,
affordances of platforms worked. They illustrated the Juan, a 20-year-old student, explained this in an insight-
role of algorithms by referring to playlists on Spotify ful manner:
and recommended items on their Netflix profile. Thus,
I understood this because of YouTube and its algor-
single-platform users considered recommendations pri- ithm. That’s how I first made contact with the word
marily as a constitutive element of the platform inter- ‘algorithm.’ People then started to explain to me more
face: a given of the platform rather than a puzzle they about it: ‘If you click on a video on YouTube about
needed to or could resolve. Accordingly, for the most dogs, it will display more videos like that one.’ So I
part they did not elaborate on how platforms worked said, ‘Something similar happens to me on Spotify
and on Netflix.’ It must be the same algorithm or
to produce such recommendations or how users could
system.
influence these suggestions. During the interviews,
users expressed this lack of awareness through state- This person’s ‘initial algorithmic experience’ was
ments such as: ‘I don’t know very well how Netflix formed by using one platform (YouTube) and exogen-
does it [to recommend content to them]’ or ‘I guess it ous factors such as other people’s interpretations of its
is because I finished watching something.’ As this algorithmic operations that gratified a cognitive need
quote suggests, answers focused often on the relation- for explanations. The combination of these two sources
ship between recommendations and the content they led to the development of one specific folk theory: algo-
usually watched or listened to on platforms. rithmic recommendations are based primarily on con-
Multi-platform users offered relatively more detailed tent similarity.
explanations about how algorithms functioned. They
often incorporated into their discussions issues that
4.2. Practical actions
reflected an understanding of algorithms as relatively
more complex entities, such as how algorithms were Regarding the actions that users carry out to obtain rec-
involved in interpreting the duration of content con- ommendations, our analysis indicates that multi-plat-
sumption, music tone analysis, and longitudinal user form users performed significantly more actions (M =
behaviour (that is, monitoring how the taste of users 0.48) compared to single-platform users (M = 0.25),
evolved over time). That was the case of Sofía, a 30- W = 9288, p = < .001, r = .22.
year-old unemployed woman, who explained: ‘I imagine Evidence from the interviews allows us to elaborate
that Netflix and Spotify take into account what is played on how the use of multiple platforms led to higher levels
in its entirety or the song that one listens in its entirety.’ of awareness of algorithms. When algorithmic rec-
Like Sofía, multi-platform users thought that the ommendations did not fit their preferences, single-plat-
‘opaque’ nature of algorithms hid a variety of capacities. form users said they tended to ignore them. Ignoring
This was illustrated by a 26-year-old graduate student recommendations was usually not carried out to influ-
who said during the interview: ‘A movie already has cat- ence algorithms but rather as a way to explore other rec-
egory markers. If I watched that movie, then Netflix is ommendations. Multiple-platform users interpreted
possibly already classifying me through those markers ‘failed’ recommendations somewhat differently. For
[…] I suppose that Spotify and other platforms will do them, practical actions were needed to shape these sug-
something similar.’ In this example, using multiple plat- gestions. Users emphasised issues of agency to account
forms is employed as evidence to support the theory that for this difference: their relationship with algorithms
algorithms can conduct several simultaneous tasks, such was an opportunity to exert their action capacity and
as content classification, user behaviour monitoring, simultaneously prevent platforms from controlling
and content recommendation based on profiling. their consumption behaviour.
For multi-platform users, awareness was a partial Regarding how they sought to exert agency, multi-
result of comparing their experiences with various algo- platform users explained the importance of indicating
rithmic platforms. As noted previously, Costa Rica typi- what content they preferred in their interactions with
cally ranks high in the use of social media in Latin algorithms. Using multiple platforms is crucial for
America, which creates fertile grounds for the adoption developing awareness in that it allows users to assess
of many platforms. In this context, Costa Rican users the efficacy of their own actions for achieving desirable
naturalised their experience with one platform by results by comparing this experience with other cases.
using it as a segue into practices with other platforms. Participants engaged in such practices by comparing
 BEHAVIOUR & INFORMATION TECHNOLOGY 7

platform affordances that seemed to them to accomplish algorithms, users assigned to algorithms certain
similar roles (such as lists on Netflix and playlists on capacities to act. Olivia, a 65-year-old retiree, thus inter-
Spotify, or favoriting content on both platforms). The preted the agency of algorithms as a form of ‘impo-
result of this comparative work informed their opinions sition,’ an attempt to limit or control her own
about platforms. Daniela, a 24-year-old university stu- autonomy. In her words: ‘I am very stubborn, I do not
dent, explained: ‘On Netflix, I don’t really know how like things imposed on me. Sometimes the recommen-
recommendations appear. I guess it is based on the dations are irrelevant, and you must skip them, you
shows you have watched entirely. On the other hand, don’t have to pay attention to them.’ For Olivia, aware-
I do like how this [recommendations] works on Spotify.’ ness was necessary to prevent this from occurring.
Building on their comparison between platforms, par- Accordingly, she adopted actions to let platforms
ticipants such as Daniela said they could better under- know she was in charge of their interaction. Olivia
stand various aspects of how algorithms work, which said she would watch again content she had already
in turn informed her appreciation of certain platforms. watched on Netflix to reaffirm her agency.
(We examine this issue in more detail in the next
section.)
4.3. Attitudes
Multi-platform users also indicated how careful they
needed to be to avoid algorithms from getting ‘confused’ To test our third hypothesis, we conducted a multiple
(as interviews put it.) For example, algorithms could linear regression analysis with user type (single-plat-
interpret that watching or listening to one show or form users coded as 0 and multi-platform users coded
song could be considered as a sign of preference for as 1), awareness, affective needs, cognitive needs, inte-
one broader category of content. In this view, agency grative needs, gender (male coded as 0 and female
needed to be enacted to let algorithms ‘understand’ coded as 1), and age as predictors of participants’ atti-
their real preferences (the term comes once again tudes (see SM, Table 2 and Figure 2). The overall
from interviewees.) One way in which they exerted regression was statistically significant (R 2 = 0.32, F(7,
this agency was by limiting who could interact with 250) = 17.96, p < .001). However, contrary to our predic-
their profiles on platforms. In their words, this helped tion, user type did not predict participants’ attitudes
prevent algorithms from getting ‘contaminated’ with towards algorithmic recommendations (β = −0.10, p
trends that did not reflect their desires. This could be = .34). Instead, the level of awareness appeared to be a
interpreted as an instance of ‘rhetorical awareness’ in better predictor (β = 0.37, p < .001). Higher levels of
that it reveals an understanding of the dynamics that awareness were associated with more positive attitudes
mutually shape the actions of both users and algorithms. towards algorithms. Furthermore, higher scores of
The use of terms such as ‘confuse’ and ‘understand’ to affective needs (β = 0.19, p < .01) and cognitive needs
describe the actions of algorithms also reveals the cen- (β = 0.16, p < .01) were associated with more positive
trality of platform personification dynamics as a mech- attitudes towards the recommendations.
anism to form awareness in Costa Rican society. It was Evidence from the interviews allows us to better
common for interviewees to refer to platforms as understand the links between user types, awareness,
human-like entities, which helped to naturalise certain and attitudes towards algorithms. Multi-platform use
actions to relate to them (Siles 2023). informed the attitudes that users had towards algor-
By being aware of them, users said they could ‘exper- ithms in two important ways. First, awareness emerged
iment’ with platforms’ algorithms. Thus, Julio, a 27- through an assessment of the individual experiences
year-old psychologist, noted he would ‘let’ Spotify’s with multiple algorithmic platforms. This led to positive
algorithms recommend content for the sake of curiosity. attitudes towards those platforms that worked best.
The use of notions such as ‘letting’ algorithms act Second, users aggregated individual experiences with
reveals how users’ awareness is tied to a sense of being multiple platforms into an overall assessment of their
in control of situations and interactions with algor- media ecologies rather than specific artifacts.
ithms. It was common for users to illustrate this by Regarding the first dynamic, users explained during
employing the language of ‘training’ to refer to how the interviews that they typically compared their indi-
they sought to ‘let’ algorithms learn what type of content vidual experiences with platforms. In this way, they
they like. developed more awareness of algorithms, which led to
In parallel, multi-platform users noted that their positive attitudes towards platforms that seemed to cap-
actions to shape platforms were also motivated by a ture their preferences best. Users’ evaluations of plat-
desire to fight the agency they attributed to algorithms forms’ strengths and limitations to recommend
themselves. As part of their rhetorical awareness of content was tied to catalog availability issues. The size
8 J. ESPINOZA-ROJAS ET AL.

of catalougs and the availability of content in the down to find something I like [on Netflix]. Maybe in the
country is crucial for Costa Ricans. As our previous beginning, when I had just created an account, it would
research has shown, Costa Ricans tend to interpret cat- give me a [random] recommendation. But now they
alog differences with the global north as a form of exclu- [Netflix and Spotify] give me things to my liking.’ Mar-
sion (Siles et al. 2019). Since they pay the same prices for ía’s assessment thus grouped together both platforms to
platforms such as Netflix and Spotify, they expect to convey her sense of satisfaction with algorithms in
have access to the same content available in countries, general.
most notably the United States. This expectation is Users often compared Netflix and Spotify with what
explained by the fact that Costa Ricans often employ they referred to as ‘social media’ more broadly (which
products such as television series, movies, and music included platforms such as Facebook and Instagram).
as means to participate in global conversations about For instance, Marco, a chemist who is 27 years old,
culture. It is thus not surprising that evaluations of the explained this difference thusly:
size of catalougs in certain platforms were tied to
Whereas [Netflix and Spotify] are focused on improv-
more positive attitudes towards algorithmic ing my experience, social media are focused on selling.
recommendations. Whereas the algorithms of social media want to obtain
Multi-platform users often considered that Spotify information from me to understand me as an individual
could make more appropriate recommendations than consumer, the algorithms of Netflix and Spotify [want
Netflix because it had a larger catalog. For that reason, to] give me something to entertain myself.
they argued, algorithms had more options to rec- Marco thus included Netflix and Spotify into one single
ommend in the first place. According to various sources category that he compared to other media technologies.
(Moody 2018), Netflix’s catalog in Costa Rica is almost The underlying principle that allowed him to group
30% smaller than in the United States. Consistent with these platforms together was their capacity to satisfy
this difference, users believed that Netflix’s small catalog affective needs.
for countries such as Costa Rica increased the possibility Despite their positive attitudes, multi-platform users
of failures. Laura, a 19-year-old college student, thus were also critical of Netflix and Spotify. Their most
explained: common complaint centred on what they perceived to
I think that Spotify has more content, it has millions of be biases in the operation of algorithmic recommen-
artists and songs. What I’ve noticed is that Spotify can dations. Some interviewees defined themselves as
give more detailed recommendations based on my ‘angered’ by this issue. Mateo, who is 30 years old and
tastes. Netflix is a little bit more limited. When I con- works in the tourism industry, thus wondered: ‘How
sider Netflix’s recommendations, I don’t find as many is it possible that [Spotify] is recommending to me
interesting things as I do when I go to Spotify.
music that I don’t listen to?’ He then provided an
Laura’s comment blended affective and cognitive needs answer to his own question: ‘These are recommen-
in her awareness of algorithmic operations. She com- dations to listen to the artists who have more streams,
bined her knowledge about the size of content catalougs more views on YouTube. They give me a screen full of
in Costa Rica to explain why one algorithmic platform recommendations for music that I won’t consume
seemed better equipped than others to capture her taste. [just because they need to honour those deals]. I feel
This dynamic was supplemented with a parallel pro- this is a waste of time.’ Positive attitudes thus shifted
cess: users aggregated these experiences to evaluate not when Mateo felt that an underlying but unstated con-
specific platforms but rather their entire media ecol- tract in the relationship between users and algorithms
ogies. On such occasions, their awareness of the features was broken.
of media ecologies permeated the attitudes towards all
the cases of those ecologies. It was common for intervie-
wees to speak of a ‘content universe’ formed by the cul-
4.4. Emotional arousal
tural products available in all the platforms they used.
Through this process, the apparent deficiencies of one To test our last hypothesis, we conducted a multiple lin-
platform were compensated for by the virtues of another ear regression analysis with user type (single-platform
platform. Users’ positive attitudes towards algorithms users coded as 0 and multi-platform users coded as 1),
stemmed from the satisfaction derived from the whole awareness, affective needs, cognitive needs, integrative
rather than its parts. For instance, María, a 23-year- needs, gender (male coded as 0 and female coded as
old university student indicated: ‘The algorithm has 1), and age as predictors of participants’ emotional
worked very well for me. I don’t have to keep scrolling arousal (see SM, Table 3 and Figure 3). The overall
 BEHAVIOUR & INFORMATION TECHNOLOGY 9

regression was statistically significant (R 2 = 0.48, F(8, gives me hours [of music]. Today I said to myself:
249) = 30.71, p < .001). The interaction between user ‘I’m happy, so I’ll put on some hip hop and some
type and awareness was statistically significant (β = reggaeton.’ But I know that I’ll play something else
later on when I go to classes.
−1.38, p < .001) and showed that for lower and medium
levels of awareness, single – and multi-platform users Multi-platform users said they were excited about rec-
significantly differed in their scores of emotional arou- ommendations that surprised them positively. This
sal. However, that was not the case for higher levels of assessment built on a form of awareness about the gen-
awareness (see SM, Table 4). Moreover, higher scores eral types of recommendations users could expect from
in affective needs (β = 0.21, p < .001) and cognitive each specific platform. Victoria, a 31-year-old project
needs (β = 0.26, p < .01) were associated with higher manager in the public sector, noted: ‘I don’t feel excited
levels of emotional arousal. about [Netflix’s] recommendations. I see them as some-
During the interviews, single-platforms users thing normal. Netflix suggests things that are similar to
reported less enthusiasm in their interactions with and what I have already watched.’ Victoria’s assessment
their uses of streaming services. In their explanations offers a glimpse into how awareness was tied to
of their relationship with these platforms, single-plat- emotional arousal: her lack of enthusiasm was explained
form users tended to downplay issues of emotion. Con- by her alleged capacity to anticipate what algorithms
versely, the role of affective needs was abundantly clear would recommend to her on Netflix.
in the explanations of multi-platform users. These users Alternatively, users said they were excited about
often tied Spotify to the experience of individual algorithms when they offered them recommendations
emotions and conceived of Netflix as an opportunity that were appropriate for the affective context that sur-
to discuss their emotions with peers. To foster this social rounded the use of platforms. Users expected the use of
experience, some interviewees said they often consumed platforms to help them cultivate certain moods and
the platform simultaneously with other people. In the emotions for those specific contexts (Siles et al. 2019).
words of Julio, a 27-year-old psychologist: Fran, a 42-year-old lawyer, provided a vivid illustration
of this practice:
I usually ask my girlfriend to watch something together
[on Netflix], using some software. We press ‘play’ at the There were about two weeks when I was very melan-
same time. [On Spotify] I just send a person a link to a cholic and [Spotify’s] algorithm recommended to me
song. These are [different] ways of sharing. One is digi- a playlist called ‘Classical Music Friday’ [that included]
tal [Spotify] and the other one is ‘experiential’ [Netflix]. rather melancholic or depressive music. I was like,
One is about human experience and the other one is ‘What the hell!’ I never expected that the algorithm
more about sharing a link. could identify that the music I was looking for was
quite depressing or quite sad!
For Julio, multi-platform comparison is crucial for
understanding the meaning of ‘sharing’ and the signifi- Fran’s words reveal two layers of emotions in his use of
cance of affective and integrative needs, such as discuss- Spotify. First, affective needs motivated his use of the
ing with others about the feelings derived from certain platform (hence his use of terms such as ‘depressive’,
recommendations. Context is once again key to situate ‘melancholic,’ and ‘sad’ to characterise his affective
this finding. In Costa Rica, a great deal of importance state). Second, his discovery of Spotify’s capacity to
is given to ‘group affiliation (as opposed to personal offer something appropriate for his affective state was
achievement)’ (Rodríguez-Arauz et al. 2013, 49). This met with strong emotions of pleasant surprise.
ends up shaping people’s relationship with algorithms
in that platforms are seen as ways to establish interper-
sonal bonds and negotiate group membership (Siles 5. Discussion
et al. 2020). When accounting for algorithmic awareness, most
Users turned to specific platform features for research has concentrated on the importance of socio-
emotional arousal. This was the case of the ‘My List’ demographic variables (Gran, Booth, and Bucher
affordance on Netflix, or the ‘Playlists’ tab on Spotify. 2021). As a supplement, in this paper we demonstrated
Irene, a 19-year-old university student, turned to Spotify the significance of using various platforms in developing
specifically for the possibilities it offered her to work on an awareness of algorithms. In short, we showed that
her emotions: multi-platform users are more aware of algorithms
I have tons of playlists because I shift from genre to than single-platform users. This finding, which supports
genre based on my moods throughout the day. I open H1, stressed the importance of analyzing users’ relation-
playlists and search for one kind of music, which ship with algorithms as part of ‘polymedia’ or broader
10 J. ESPINOZA-ROJAS ET AL.

‘digital environments’ (Boczkowski and Mitchelstein features on the interface). In contrast, multi-platform
2021; Madianou and Miller 2013). users also incorporated numerous exogenous factors
In more qualitative terms, it could be argued that (such as context variables that included their moods
multi-platform and single-platform usage is tied to or an interpretation of platforms’ commercial inten-
different types of awareness. Explanations given by tions). Thus, our study expanded previous findings by
single-platform users revealed for the most part more showing that users consistently mobilise insights from
basic forms of awareness (Koenig 2020): users under- their experiences with both endogenous and exogenous
stood that algorithms were performing certain actions factors, which profoundly shapes their behaviours with
in order to recommend them personalised content but and expectations of algorithmic platforms.
did not inquire beyond this basic premise. Alternatively, H3 stated that multi-platform users have more posi-
multi-platform users typically provided explanations tive attitudes towards algorithmic recommendations
that were closer to ‘rhetorical awareness’: they revealed than single-platform users. However, data from our
an understanding of the ‘mutual domestication’ study indicate that awareness was a better predictor
dynamics involved in their relationship with algorithms than user type. In the United States, studies show that
(Siles et al. 2019). users typically assess the intervention of algorithms in
Our findings also confirmed the significance of age in daily life with concern and skepticism (Smith 2018).
accounting for awareness. According to Gran, Booth, This concern comes primarily from the belief that algor-
and Bucher (2021), there is a ‘generational divide in ithms always express human biases. To be sure, some
algorithm awareness. ‘No awareness’ of algorithms is users in our study expressed similar concerns about
highest among the older respondents, while the two the operation of algorithms in curating their experiences
highest level of awareness is found among the youngest with platforms. There was also a preoccupation with the
age groups’ (1785). Although our sample of multi-plat- possibility that algorithms were preventing them from
form users was relatively younger than single-platform discovering new content that they could potentially
users, our analysis showed that this ‘generational divide’ enjoy. Yet, for the most part, those who were more
could be partially explained by the prevalence of multi- aware of algorithms tended to assess positively the rec-
platform use among younger age groups. ommendations they received on Netflix and Spotify.
Not only were multi-platform users more aware of These attitudes might come from the sense that algor-
algorithms than single-platform users but they also ithms are responding primarily to the actions users
adopted their behaviour accordingly. The users we have undertaken to personalise content. In other
studied often considered the implications of algorithms words, users appreciated content that signalled that
and focused on how adjustments on both their actions algorithmic personalisation was taking place (c.f.
and algorithmic operations could yield better rec- Swart 2021).
ommendations. They engaged in more practices to Finally, our analysis also showed an interaction effect
influence algorithmic recommendations than single- between user type and awareness, which explains levels
platform users (thus supporting H2). User actions also of emotional arousal (H4). This brings to the fore the
included attempts to ‘fight’ what users perceived to be affective nature of the affective relationship between
algorithmic ‘impositions.’ Dogruel, Facciorusso, and users and algorithms. Users assess the role of algorithms
Stark (2020)’s conclusion that awareness of algorithms and the relevance of their recommendations through an
is tied to a perceived sense of autonomy in users could emotional lens that privileges how either movies or
help explain this finding. music allow them to nurture or cultivate emotions
Previous studies have identified relatively similar and moods. The use of various platforms was crucial
dynamics. Swart (2021) showed that users typically in this process as it allows people to satisfy affective, cog-
compare different websites to make sense of algorithmic nitive, and integrative needs.
curation. Considering folk theories of algorithmic rec-
ommendations on Spotify, Siles and colleagues (2020)
6. Concluding remarks
argued that users’ understanding of this platform
stemmed partially ‘from how they conceive[d] of infra- Our analysis focused on two platforms in the domain of
structures and technologies other than this platform. entertainment. Future research could shed light on how
Users typically incorporate[d] their experiences with awareness varies when more artifacts are analyzed or
other algorithmic devices into their understanding of when considering practices and attitudes in areas
Spotify’ (5). In our study, single-platform users pro- other than culture and entertainment. It could also be
vided explanations of algorithms that focused almost of interest to examine the experiences of users who
entirely on endogenous platform elements (such as have distinct kinds of ties with algorithms. For example,
 BEHAVIOUR & INFORMATION TECHNOLOGY 11

a potential study could examine gig workers’ algorith- ORCID

mic awareness when interacting with a variety of apps Johan Espinoza-Rojas http://orcid.org/0000-0001-8292-
in delivery services or transportation. 8512
Our analyses need to be situated within the specific Ignacio Siles http://orcid.org/0000-0002-9725-8694
context of Costa Rican society. We showed that the Thomas Castelain http://orcid.org/0000-0001-9415-1771
prevalence of social media in Costa Rica created fertile
grounds for users to compare their experiences with var-
ious platforms and thus form awareness of algorithms. References
We also considered how the tendency to personify algo- Anderson, Adam. 2005. “Affective Influences on the
rithmic platforms played a part in the number of practical Attentional Dynamics Supporting Awareness.” Journal of
actions that multi-platform users undertook in their Experimental Psychology: General 134 (2): 258–281.
doi:10.1037/0096-3445.134.2.258.
engagement with these platforms. The size of platform Boczkowski, Pablo, and Eugenia Mitchelstein. 2021. The
catalougs and the availability of content mattered for Digital Environment: How We Live, Learn, Work, and
Costa Ricans in ways that perhaps differ from users in Play Now. Cambridge, Massachusetts: MIT Press.
other countries. This is because people in Costa Rica Bodó, Balázs, Natali Helberger, Sarah Eskens, and Judith
tend to use products such as movies and music to partici- Möller. 2019. “Interested in Diversity.” Digital Journalism
7 (2): 206–229. doi:10.1080/21670811.2018.1521292.
pate in global conversations about culture and thus feel
Bucher, Taina. 2017. “The Algorithmic Imaginary: Exploring
excluded when they have no access to certain content the Ordinary Affects of Facebook Algorithms.”
(despite paying the same rates as in the global north.) Information, Communication & Society 20 (1): 30–44.
This ended up shaping awareness of algorithms in impor- doi:10.1080/1369118X.2016.1154086.
tant ways. Finally, the contextual importance of group Bucher, Taina. 2018. If … Then: Algorithmic Power and
membership over personal achievement in this country Politics. Oxford: Oxford University Press.
Büchi, Moritz, Eduard Fosch-Villaronga, Christoph Lutz,
was tied to the fulfilment of integrative needs and Aurelia Tamò-Larrieux, and Shruthi Velidi. 2021.
emotional arousal derived from algorithmic platforms. “Making Sense of Algorithmic Profiling: User Perceptions
Novel studies could compare how awareness of algor- on Facebook.” Information, Communication & Society.
ithms varies in other contexts both in the global south doi:10.1080/1369118X.2021.1989011.
and other parts of the world. Burrell, Jenna. 2016. “How the Machine ‘Thinks’:
Understanding Opacity in Machine Learning
One of the contributions of our study centres on the
Algorithms.” Big Data & Society 3 (1): 205395171562251–
use of mixed methods to account for users’ awareness of 12. doi:10.1177/2053951715622512.
algorithms. This methodological approach allowed us to Christin, Angèle. 2020. “The Ethnographer and the
understand the similarities and differences in percep- Algorithm: Beyond the Black box.” Theory and Society 49
tion and behaviour between multi-platform and (5–6): 897–918. doi:10.1007/s11186-020-09411-3.
single-platform users, while also accounting for the Corporación Latinobarómetro. 2018. Latinobarómetro.
Santiago: Corporación Latinobarómetro.
nuances of their explanations in a detailed manner. Deng, Shuiguang, Dongjing Wang, Xitong Li, and Guandong
However, because of our sampling strategy, our respon- Xu. 2015. “Exploring User Emotion in Microblogs for
dents were primarily college students who lived near Music Recommendation.” Expert Systems with Applications
Costa Rica’s capital. We sought to balance out this 42 (23): 9284–9293. doi:10.1016/j.eswa.2015.08.029.
bias during the second phase of our study by seeking DeVito, Michael, Jeremy Birnholtz, Jeffery Hancock, Megan
French, and Sunny Liu. 2018. “How People Form Folk
specifically to interview users with different sociodemo-
Theories of Social Media Feeds and What it Means for
graphic characteristics and educational levels. Future How We Study Self-Presentation.” Paper presented at the
research could focus on much more representative 2018 CHI Conference on Human Factors in Computing
samples at a national or regional level. systems, Montreal, QC, Canada, April 21-26, 1–12.
doi:10.1145/3173574.3173694.
DeVito, Michael, Darren Gergle, and Jeremy Birnholtz. 2017.
Acknowledgements ““Algorithms Ruin Everything”: #riptwitter, Folk Theories,
and Resistance to Algorithmic Change in Social Media.” Paper
We thank Carlos Brenes-Peralta and Larissa Tristán-Jiménez presented at the 2017 CHI Conference on Human Factors in
for their help in developing the arguments of this paper. We Computing systems, Denver, Colorado, May 6-11, 3163–3174.
also thank the anonymous reviewers for their most helpful Dogruel, Leyla, Dominique Facciorusso, and Birgit Stark.
suggestions. 2020. “‘I’m Still the Master of the Machine.’ Internet
Users’ Awareness of Algorithmic Decision-Making and
Their Perception of its Effect on Their Autonomy.”
Disclosure statement Information, Communication & Society. Published electro-
nically December 30, 2020, doi:10.1080/1369118X.2020.
No potential conflict of interest was reported by the author(s). 1863999.
12 J. ESPINOZA-ROJAS ET AL.

Dourish, Paul, and Victoria Bellotti. 1992. “Awareness and Red 506. 2018. Red 506. San José: El Financiero.
Coordination in Shared Workspaces.” Paper presented at the Rodríguez-Arauz, Gloriana, Marisa Mealy, Vanessa Smith, and
1992 ACM Conference on computer-supported cooperative Joanne DiPlacido. 2013. “Sexual Behavior in Costa Rica and
work, Toronto, Ontario, Canada, November 1-4, 107–114. the United States.” International Journal of Intercultural
Eriksson, Maria, Rasmus Fleischer, Anna Johansson, Pelle Relations 37 (1): 48–57. doi:10.1016/j.ijintrel.2012.06.002.
Snickars, and Patrick Vonderau. 2019. Spotify Teardown: Schwartz, Sander, and Martina Mahnke. 2021. “Facebook use
Inside the Black box of Streaming Music. MIT Press. as a Communicative Relation: Exploring the Relation
Eslami, Motahhare, Aimee Rickman, Kristen Vaccaro, Between Facebook Users and the Algorithmic News
Amirhossein Aleyasen, Andy Vuong, Karrie Karahalios, Feed.” Information, Communication & Society 24 (7):
Kevin Hamilton, and Christian Sandvig. 2015. ““I Always 1041–1056. doi:10.1080/1369118X.2020.1718179.
Assumed That I Wasn’t Really That Close to [her]”: Sheth, Bhavin, and Thuan Pham. 2008. “How Emotional
Reasoning About Invisible Algorithms in News Feeds.” Arousal and Valence Influence Access to Awareness.”
Paper presented at the 33rd annual ACM Conference on Vision Research 48 (23-24): 2415–2424. doi:10.1016/j.
Human Factors in Computing systems, Seoul, Republic of visres.2008.07.013.
Korea, April 18-23, 153–162. Siles, Ignacio. 2023. Living with Algorithms: Agency and User
Eslami, Motahhare, Kristen Vaccaro, Min Kyung Lee, Amit Culture in Costa Rica. Cambridge, Massachusetts: MIT Press.
Elazari Bar On, Eric Gilbert, and Karrie Karahalios. 2019. Siles, Ignacio, Johan Espinoza-Rojas, Adrián Naranjo, and
“User Attitudes Towards Algorithmic Opacity and María Fernanda Tristán. 2019a. “The Mutual
Transparency in Online Reviewing Platforms.” Paper pre- Domestication of Users and Algorithmic
sented at the 2019 CHI Conference on Human Factors in Recommendations on Netflix.” Communication, Culture
Computing systems, Glasgow, Scotland UK, May 4-9, 1–14. & Critique 12 (4): 499–518.
Gran, Anne-Britt, Peter Booth, and Taina Bucher. 2021. “To Siles, Ignacio, and Ariana Meléndez-Moran. 2021. “The Most
be or not to be Algorithm Aware: A Question of a new Aggressive of Algorithms”: User Awareness of and
Digital Divide?” Information, Communication & Society Attachment to TikTok’s Content Personalization.” Paper
24 (12): 1779–1796. doi:10.1080/1369118X.2020.1736124. presented at the 71st annual conference of the
Grzymek, Viktoria, and Michael Puntschuh. 2019. What International Communication association, Denver, 2021.
Europe Knows and Thinks About Algorithms: Results of a Siles, Ignacio, Andrés Segura-Castillo, Mónica Sancho, and
Representative Survey. Gütersloh: Bertelsmann Stiftung. Ricardo Solís-Quesada. 2019b. “Genres as Social Affect:
Hussain, Ashfaq, Ghulam Shabir, and H. Taimoor Ul. 2020. Cultivating Moods and Emotions Through Playlists on
“Cognitive Needs and use of Social Media: A Spotify.” Social Media + Society 5 (2): 1–11.
Comparative Study of Gratifications Sought and Siles, Ignacio, Andrés Segura-Castillo, Ricardo Solís, and
Gratification Obtained.” Information Discovery and Mónica Sancho. 2020. “Toward Multi-Label Sentiment
Delivery 48 (2): 79–90. doi:10.1108/IDD-11-2019-0081. Analysis: A Transfer Learning Based Approach.” Journal
Instituto Nacional de Estadística y Censo. 2020. Encuesta of Big Data 7 (1): 1–15.
Nacional de Hogares 2020. San José: Instituto Nacional Sjöblom, Max. 2015. “Watching others play: A uses and gra-
de Estadística y Censo (INEC). https://www.inec.cr/ tifications approach to video game streaming motives”.
flipbook/enaho2020/index.html?web=1&wdLOR= Master’s thesis, Aalto University. Aaltodoc. https://
c3B556A6D-DCB9-A447-AF3E-F3C3AFA6F2C1. aaltodoc.aalto.fi/handle/123456789/18171.
Koenig, Abby. 2020. “The Algorithms Know Me and I Know Smith, Aaron. November 16, 2018. “Public Attitudes Toward
Them: Using Student Journals to Uncover Algorithmic Computer Algorithms”. Pew Research Center: Internet,
Literacy Awareness.” Computers and Composition 58: Science & Tech. https://www.pewresearch.org/internet/
102611–14. doi:10.1016/j.compcom.2020.102611. 2018/11/16/public-attitudes-toward-computer-algorithms/.
Lobato, Ramon. 2019. Netflix Nations. New York University Swart, Joëlle. 2021. “Experiencing Algorithms: How Young
Press. People Understand, Feel About, and Engage With
Madianou, Mirca, and Daniel Miller. 2013. “Polymedia: Algorithmic News Selection on Social Media.” Social
Towards a new Theory of Digital Media in Interpersonal Media + Society 7 (2): 205630512110088–11. doi:10.1177/
Communication.” International Journal of Cultural 20563051211008828.
Studies 16 (2): 169–187. doi:10.1177/1367877912452486. Tefertiller, Alec. 2017. “Moviegoing in the Netflix Age:
Mason, Jennifer. 2006a. Six strategies for mixing methods and Gratifications, Planned Behavior, and Theatrical
linking data in social science research. Research output: Attendance.” Communication & Society 30 (4): 27–43.
Working paper. National Centre for Research Methods Tefertiller, Alec. 2018. “Media Substitution in Cable Cord-
Working Papers. Cutting: The Adoption of Web-Streaming Television.”
Mason, Jennifer. 2006b. “Mixing Methods in a Qualitatively Journal of Broadcasting & Electronic Media 62 (3): 390–
Driven way.” Qualitative Research 6 (1): 9–25. 407. doi:10.1080/08838151.2018.1451868.
Moody, Rebecca. January 20, 2018. “Which countries pay the Thurman, Neil, Judith Moeller, Natali Helberger, and Damian
most and least for Netflix?” www.comparitech.com. Trilling. 2019. “My Friends, Editors, Algorithms, and I.”
Pasquale, F. 2015. The Black Box Society: The Secret Digital Journalism 7 (4): 447–469. doi:10.1080/21670811.
Algorithms That Control Money and Information. 2018.1493936.
Cambridge, Massachusetts: Harvard University Press. Ytre-Arne, Brita, and Hallvard Moe. 2021. “Folk Theories of
Powers, Elia. 2017. “My News Feed is Filtered?” Digital Algorithms: Understanding Digital Irritation.” Media,
Journalism 5 (10): 1315–1335. doi:10.1080/21670811.2017. Culture & Society 43 (5): 807–824. doi:10.1177/
1286943. 0163443720972314.