See discussions, stats, and author profiles for this publication at: https://www.researchgate.

net/publication/381109399

Debunking Instagram's Algorithm-Sugarcoating

Article in SSRN Electronic Journal · May 2024

DOI: 10.2139/ssrn.4851736

CITATIONS READS

0 299

4 authors:

Tobey Gross André Michaud

SRH Hochschule Heidelberg Service de Recherche Pédagogique SRP Inc
30 PUBLICATIONS 2 CITATIONS 220 PUBLICATIONS 1,967 CITATIONS

SEE PROFILE SEE PROFILE

Asaad Chasib Hamood Yassine Zerrouki

University of Basrah 3 PUBLICATIONS 0 CITATIONS
30 PUBLICATIONS 5 CITATIONS
SEE PROFILE
SEE PROFILE

All content following this page was uploaded by Tobey Gross on 03 June 2024.

The user has requested enhancement of the downloaded file.

Zentrum für
Medienpsychologie und
Open Access
Verhaltensforschung CC BY 4.0

Citation: Gross, T., Michaud, A., Zerrouki, Y., Hamood, A. (2024). Debunking Instagram's Algorithm-Sugarcoating.
Zentrum für Medienpsychologie und Verhaltensforschung, 05/2024.

Corresponding Tobey Gross1, André Michaud2, Yassine Zerrouki3, Asaad Hamood4

Authors:

Contact Information: [email protected]

Type: Sentiment Analysis, Review of Literature

Field: Psychology, Economics, Information Technology Research Funding: none

Date: May 31, 2024 Conflict of Interest: 5

1 ZeMV
2Service de Recherche
Pedagogique Inc.
3 Mohammed I University,

Oujda, IDS
4 Basrah University

This work is licensed under Creative Commons Attribution 4.0 International.

Abstract

In 2021 and 2023, there were two distinct official statements published on the
mechanics behind the algorithm responsible for Instagram's content feed curation,
where the 2023 publication, named "Instagram Ranking Explained" is presented
as an updated and expanded version of the one in 2021, named "Shedding More
Light on How Instagram Works". This paper examines the statements made
therein by Adam Mosseri, the head of Instagram who authored the publications,
comparing them with insights from contemporary literature and investigating the
sentiment of the publications through a mixed-method sentiment analysis. The
analysis aims to show, that the statements present the algorithm in a particularly
positive light, downplaying and largely ignoring potential detriments. Statements
are examined for pseudo-transparency, providing a veneer of openness, while
concealing deeper economic motives through obscuring practices like data
extraction and engagement maximization, especially in mid of rising criticism
towards social networking sites.

Keywords: Algorithmic Personalization, Digital Media, Social Media, Content

Curation, Framing, Sentiment Analysis, Recommender Algorithms

ZeMV e-Publikation 05/2024 1 https://zemv.org

Zentrum für
Medienpsychologie und Gross, Michaud, Zerrouki, & Hamood
Verhaltensforschung

Introduction

Instagram assumes a major role in the world of social media. It manages to sustain users'
interest and attention in manifold ways, one of which is showing those sorts of contents, that
are most engaging. But in what ways does Instagram determine what contents to display to
individuals? What makes these pictures or videos so special that they can capture attention
reliably, on a large scale and perpetually? The answer lies within an algorithm that controls
what people see when they log into their accounts and the order in which contens are ranked.

What sounds relatively unsuspicious at first, and seems to be a likeable feature for navigating
through endless arrays of contents, has grown to a major concern in the global tech industry,
perhaps the most worrying issue in modern digital economy overall. While algorithmic
content curation is indeed a sophisticated manner of tailoring user experiences, it raises
fundamental questions about transparency, corporate motives and depth of data mining and
surveillance. Because in reality, users are subjected to extensive surveillance, tracking and
recording every interaction. The depth of data extraction is staggering, and in a fast-paced
digital ecosystem, behavioral data functions as a commodity; enabling big data companies to
craft disturbingly accurate profiles of each individual, that are sold to the highest bidder and
used to predict and carefully manipulate behavior.

It is still highly challenging for scientists to unveil the extent of these sinister motives behind
seemingly benign digital architectures, because none of the big enterprises behind those
algorithms are willing to disclose their data, practices or mechanisms – for obvious reasons.
However, the ethical question about what many nowadays refer to as Surveillance Capitalism
has apparently a unison answer: users are being exploited with little regard to digital
autonomy, ethics and consequences.

All the more, communicating the opposite of what is really happening behind the curtain
might be perceived as a blatant attempt to throw concerned individuals off track and
strengthen their bonds and trust into the platform. We claim, that this is the exact reason for
two publications authored by Instagram's CEO and disclosed on their own website, at a time
when the global conversation around data protection and immoral practices in big data
enterprises is becoming louder. Mosseri tries to reassure the reader, that Instagram's users,
their experience and well-being is at the heart of what Instagram and its algorithms are after.
We uncompromisingly challenge that notion by presenting an array of avaliable research, that
exposes, how algorithmic personalization primarily serves to keep users engaged and
extracting their behavioral surplus in order to monetize it. Furthermore, the following probe
quantitatively and qualitatively examines the sentiments communicated by the 2023
publication and repositions them as sugarcoated half-truths meant to disguise and distract
from Instagram's algorithmic intentions, which, as we will find, are no different from any of
the attention economy industry's standards.

ZeMV e-Publikation 05/2024 2 https://zemv.org

Zentrum für
Medienpsychologie und Gross, Michaud, Zerrouki, & Hamood
Verhaltensforschung

Quantitative Sentiment Analysis

For the quantitative analysis, we employed three distinct methods: TextBlob, VADER and
Flair. Each of those methods offers a unique approach to sentiment analysis, with its own
emphasis and analysis methodology. Through this multifaceted approach, we obtained a
nuanced impression of the conveyed sentiment in the publication.

The methodology included downloading the full text, cleaning it and dividing it in suitable
coherent segments, that would preserve the context. Moreover, segmentation allowed for
granular insight in our analysis, providing us with the opportunity to identify shifts or
variations in sentiment. Lastly, the collection of more data points offered a richer set of
metrics to analyze, increasing validity and reliability of our analysis.

Through application of three different methods (which are explained in more detail below),
we were able to cross-validate sentiment scores and ensure consistency across different
analytical methods. In a single-method analysis, nuances in sentiment might have been
overlooked or not captured due to noise etc. Hence, a multi-method analysis appeared more
robust for our investigation.

Below, there is an overview of the results of each analysis.

TextBlob TextBlob VADER VADER VADER VADER Flair Flair

Segment
Polarity Subjectivity Negative Neutral Positive Compound Sentiment Confidence

Initial Paragraph 0.30142 0.550568 0.0 0.809 0.191 0.9358 NEGATIVE 0.968971

Second Larger Segment 0.25303 0.579545 0.0 0.843 0.157 0.9766 POSITIVE 0.653849

Third Larger Segment 0.115463 0.493416 0.029 0.823 0.148 0.9985 POSITIVE 0.553345

Fourth Larger Segment 0.1348 0.542008 0.044 0.826 0.13 0.9954 POSITIVE 0.978636

Fifth Larger Segment 0.152114 0.459408 0.029 0.794 0.177 0.9991 POSITIVE 0.778824

Final Segment 0.247598 0.46656 0.021 0.82 0.16 0.9923 NEGATIVE 0.986306

Fig. 1: Analysis Matrix

ZeMV e-Publikation 05/2024 3 https://zemv.org

Zentrum für
Medienpsychologie und Gross, Michaud, Zerrouki, & Hamood
Verhaltensforschung

Fig. 2: Bar Chart Comparison

ZeMV e-Publikation 05/2024 4 https://zemv.org

Zentrum für
Medienpsychologie und Gross, Michaud, Zerrouki, & Hamood
Verhaltensforschung

Fig. 3: Heatmap Distribution

ZeMV e-Publikation 05/2024 5 https://zemv.org

Zentrum für
Medienpsychologie und Gross, Michaud, Zerrouki, & Hamood
Verhaltensforschung

Fig. 4: Overall Sentiment Score by Segment (VADER Compound, Flair Normalized)

ZeMV e-Publikation 05/2024 6 https://zemv.org

Zentrum für
Medienpsychologie und Gross, Michaud, Zerrouki, & Hamood
Verhaltensforschung

Legend and Discussion

TextBlob is a simple yet effective tool for natural language processing tasks. Its two key
metrics are polarity and subjectivity score, where the former measures the degree of positivity
or negativity on a scale from -1.0 to 1.0, with higher values indicating a more positive
sentiment, and the latter indicates the amount of personal opinion versus factual information
on a scale from 0.0 to 1.0, with higher values indicating more subjective content.

VADER (Valence Aware Dictionary and sEntiment Reasoner) is particularly attuned to the
sentiment in social media contexts and is capable of capturing nuances, such as exclamation
marks and capitalization. VADER provides four metric scores: positive, negative, neutral and
a compound score, which calculates the aggregate of the overall sentiment into a single metric
on a scale from -1.0 (most negative) to 1.0 (most positive). Consequently, higher values in
VADER Negative indicate higher proportions of negative sentiment, higher values in VADER
Positive indicate higher proportions of positive sentiment and VADER neutral indicate higher
proportions of neither positive nor negative, but rather neutral sentiment. The aggregate of all
three sums up to 1.0.

VADER Compound score is a normalized, weighted composite score that takes into account
the proportion of negative, neutral and positive sentiment. It is thus not a direct average,
which means due to weighting and normalization in the VADER algorithm, the values of each
distinct metric are calculated against each other. Hence, in texts with predominantly neutral
sentiment, a small proportion of positive score can result in a significantly higher composite
score, if the counterweight metric (negative sentiment) is nearly absent. This way of
calculation makes VADER Compound a highly expressive metric considering both the
proportion and intensity of sentiment scores across a text.

Flair is a more sophisticated deep learning sentiment analysis tool with a binary sentiment
classification and a confidence score that indicates the certainty of the classification on a scale
from 0.0 to 1.0. In our visualization, we employed a numeric value of -1.0 for negative and
1.0 for positive for the binary classification.

ZeMV e-Publikation 05/2024 7 https://zemv.org

Zentrum für
Medienpsychologie und Gross, Michaud, Zerrouki, & Hamood
Verhaltensforschung

As for the discussion of the results, starting with TextBlob, the calculated polarity scores
exhibit consistently positive results, ranging from 0.115 to 0.301, which indicates an overall
positive sentiment. The subjectivity scores range from 0.459 to 0.58, which suggests a
relatively balanced mix of subjective and factual / neutral content throughout the publication.

VADER provided an overwhelmingly positive sentiment regarding the range of the compound
score from 0.9358 to 0.9991. While positive sentiment scores range from 0.13 to 0.191,
negative scores range from 0.0 to 0.044 and neutral scores range from 0.794 to 0.843, the
compound score factors in a weighted normalization of all three metrics, which explains that
the high value in compound is based on predominantly neutral sentiment with tendencies to
positive rather than negative throughout the publication.

The results of the analysis using Flair were slightly more nuanced. While the majority of
segments was identified as positive, with confidence levels varying between 0.553 and 0.978,
two segments were identified as negative with high confidence scores 0.969 and 0.986. This
suggests, that Flair, which is capable of contextual analysis, identified specific phrases or
contexts, that it could interpret as negative. Notably, those were the first and the last segment
of the publication. Flair uses a neural network model, which is capable of capturing context
and subtleties better, but might also be influenced by specific phrases that carry negative
sentiment, even if the overall text is positive. Since in the first segment, the author addresses
"a lot of misconceptions" and the final segment discusses "shadowbanning" (a theory
experienced very negatively by users), where the term alone carries a notion of censorship and
lack of transparency, this might deliver an explanation for Flair's negative classification.

Finally, the line plot featuring the overall sentiment score by segment (Fig. 4) is a compound
metric we calculated from VADER Compound scores and Flair confidence, both derived
directly from the sentiment analysis results. For this metric, we normalized the Flair numeric
by the sentiment type. Positive sentiment was considered 1, negative was considered -1 to
reflect the negative impact.

Thus, the general calculation is:

VADER Compound + ( Flair Confidence × Flair Sentiment Numeric)

Compound Sentiment Score =
2

ZeMV e-Publikation 05/2024 8 https://zemv.org

Zentrum für
Medienpsychologie und Gross, Michaud, Zerrouki, & Hamood
Verhaltensforschung

Example calculation for the first segment (="initial paragraph"):

VADER Compound: 0.9358

Flair Confidence: 0.968971
Flair Sentiment: Negative (converted to -1)

0.9358 + (0.968971 × −1) 0.9358 − 0.968971

Compound Sentiment Score(Segment1) = = = − 0.0165855
2 2

Justification for this metric:

Not only does this metric include the overall sentiment provided by VADER, but it also
reflects the sentiment classification by Flair and likewise, its certainty. Moreover, the
averaging process ensures that both VADER's and Flair's classifications are equally
considered, and Flair's binary classification impacts the score through normalization to 1 and
-1. Finally, the reason to choose VADER and Flair over Textblob was the latter's separation of
subjectivity and sentiment scoring, which does not directly exhibit a composite sentiment
score and is thus not as straightforward as the former two. In addition, VADER is tuned to
social media contexts, and Flair is sophisticated through its contextual sensitivity, both of
which are features that are useful to take into account in the calculation of an aggregate
overall score. The need for such integrated analyses voted against an integration of Textblob
in this particular metric.

In conclusion, our analysis of Adam Mosseri's 2023 publication shows an overarching

positive sentiment. The combination of metrics and analyses conducted provides a robust
multidimensional proof of the conveyed sentiment and solidifies the assumption, that, despite
slight variations across the publication, the content is tailored to reinforce the narrative of user
benefit and intentions to enhance users' experience on the platform, portraying the
employment of algorithmic content curation as a choice that is beneficial to individuals,
deliberately conveying a sense of positive impact and "nothing to worry about".

ZeMV e-Publikation 05/2024 9 https://zemv.org

Zentrum für
Medienpsychologie und Gross, Michaud, Zerrouki, & Hamood
Verhaltensforschung

Qualitative Sentiment Analysis

For the qualitative analysis of the publications sentiment, we systematically examined the
language and framing techniques used throughout the text.

1. Introduction with transparency claim

Quote: "We want to do a better job of explaining how Instagram works. There are a lot of
misconceptions out there, and we recognize that we can do more to help people, especially
creators, understand what we do.", "...we recognize", "...shed more light", "...response to
feedback", "...help improve the experiences"

Analysis: This opening statement frames Instagram strongly as a user-focused and transparent
platform, using phrases like "a better job of explaining" and "help people" (especially before a
comma-separation), that specifically portray a commitment to openness. Furthermore, the
term "misconceptions" subtly shifts the blame to the public, who obviously do not understand
Instagram's operations. Thereby, all further claims of any potential wrongdoing are already to
be easily subsumed under "misconceptions". Furthermore, employing the quoted highly
positively connoted phrases present Instagram, from the introduction on, as a user-centered,
user satisfaction-committed, input and criticism-acknowledging platform, that puts its
emphasis on continuous evolution based on user input and creates a welcoming and inclusive
sentiment, as expected by a transparent institution, almost framing Instagram as similar to a
non-profit entity.

2. Algorithm and Personalization

Quote: "Instagram doesn‘t have a singular algorithm that oversees what people do and don‘t
see on the app. We use a variety of algorithms, classifiers, and processes, each with its own
purpose."

Analysis: Firstly, by detailing the complexity of their systems, Instagram does appear
sophisticated and thorough, fostering a particular sense of trust. Again, the "clarification" is
introduced with the notion, that the statement is a hypothetical response to a "misconception",
since it directly opens with a negation, instead of an affirmative sentence. Apart from that, the
sentence ends, before another clarifying statement is made. Instagram doesn‘t (...) period.
This suits the demystifying nature of the statement, reducing fear and scepticism about a
"monolith" control mechanism, rather softening the public's concerns, in watering down the
algorithm-based nature of content curation to carefully selected parts, that have an
individualized experience for the user in mind, rather than a profit-oriented manipulative
nature of content curation.

ZeMV e-Publikation 05/2024 10 https://zemv.org

Zentrum für
Medienpsychologie und Gross, Michaud, Zerrouki, & Hamood
Verhaltensforschung

3. Customization and User Control

Quote: "We rank things differently in these different parts of the app, and have added features
and controls like Close Friends, Favorites and Following so you can further customize your
experience.", "We want to make the most of people's time..."

Analysis: Firstly, the emphasis on user control takes away the "purported" element of
"remote-control" by algorithmic curation and creates a fictitious verbal distance to the actual
purpose of the algorithm; instead, the reader is actively portrayed as the controlling element,
exactly mirroring the mechanism behind algorithmic curation: "the user controls the
algorithm" is the notion that is conveyed, rather than the algorithm being an actively
manipulative element to content curation in the first place. Coupled with the image of eliciting
the most out of users' time, the illustration of algorithm mechanics is further warped to reflect
a sense of not only self-controlling one's content curation, but being supported by Instagram's
implementation of algorithms. This reinforces agency and satisfaction, and likewise, it subtly
portrays the platform itself as individually useful, since it helps "making the best of one's
time" (which can be seen as the very opposite to what is commonly referred to as
"doomscrolling": the purposeless half-aware scrolling through content, while not realizing
how much time passes, often leaving the user frustrated about their own unproductivity and
waste of time).

4. Community and Safety Emphasis

Quote: "We always want to lean towards letting people express themselves, but when someone
posts something that may jeopardize another person's safety, we step in."

Analysis: This dual-focus on self-expression and safety positions Instagram as a balanced

platform that values individual freedom and community welfare. While this is a valuable
approach, there is solely the content creation concerned in safety considerations, which can
distract from the very thought, that content curation (that no user has an active influence on)
might also pose detriment to community welfare.

ZeMV e-Publikation 05/2024 11 https://zemv.org

Zentrum für
Medienpsychologie und Gross, Michaud, Zerrouki, & Hamood
Verhaltensforschung

5. Addressing Shadowbanning Concerns

Quote: "Contrary to what you might have heard, it's in our interest as a business...", "...more
we can do to increase transparency"

Analysis: Opening the statement with another apparent "misconception", the choice of words
in the paragraph is predominantly encouraging and conveys a sense of "collaboration"
between Instagram and its users. By apparently aligning Instagram's business interests with
users' success, the author creates a fictitious bond that is characterized by mutual trust and
support, which is optimal for casting doubt on claims of manipulative practices and any
"purportedly manipulative" practice.

6. Positive Word Association, Framing Techniques

Quote: "...home base", "...friends", "...family", "...closest friends", "...discover", "...we do our
best", "...you help improve the experience" etc.

Analysis: Across the publication, there is an array of very positively associated terms and
expressions, that evoke senses of comfort, safety, collaboration, openness, trust, intimacy,
emotional attachment, connection, welcoming atmospheres, excitement, adventure, fun and
joy.

All this positive vocabulary is intended to establish a highly positive association with the
platform, intentionally integrating and including the individual as part of the evolution of the
entire platform. Particularly the repeatedly reinforced sense of companionship between
Instagram and the user, where the users are empowered to influence their own experience is a
crucial linguistic tool to move the notion away from being a subject of manipulation in any
way or form. This framing technique is consistently employed across the entire publication,
continually reminding the reader of their highlighted influence and control, while diminishing
the amount of control, that the algorithm has in the curation of content.

Furthermore, the language is aimed at creating a sense of belonging, where the platform's
practices are "transparently" simplified (and, in addition, positioned alongside industry
standards, through the statement "With any ranking algorithm...", which further solidifies the
harmless, common and unsuspicious nature), suggesting openness, honesty, agency and first
and foremost, always designed with the user's best interest in mind – thereby systematically
downplaying potential conflicts of interest between the platform and its users and distracting
from the own economic interests.

ZeMV e-Publikation 05/2024 12 https://zemv.org

Zentrum für
Medienpsychologie und Gross, Michaud, Zerrouki, & Hamood
Verhaltensforschung

Conclusive Statement

Overall, the comprehensive sentiment analysis of Adam Mosseri's statements reveals a

deliberate use of positive language and framing techniques to create a user-friendly image of
Instagram as a platform, and attempts to convey transparency about its algorithmic practices.
Both our quantitative and qualitative sentiment analyses revealed a predominantly positive
sentiment, highlighting the strategic mitigation of concerns and criticisms, while
systematically appealing to users' sense of trust. This portrayal tends to obscure the platform's
economic motives and intentionally moves attention away from potentially manipulative, or
generally negative aspects of its algorithms.

In the following section, we will critically review existing literature on algorithms and content
curation, insights into industry standards on engagement maximization practices and general
scientific knowledge on attention economy. Contrasting Mosseri's statements with those
findings will provide more nuanced insights in how far the conveyed image is inaccurate and
shed more light on how Instagram works.

Literature Review

According to Mosseri's portrayal, algorithms primarily serve as a tool working solely in favor
of Instagram's users, designed to enhance the experience on the platform, by learning the
user's preferences and displaying more content directly relevant to them. As we already
established in the qualitative sentiment analysis, the publication in question thereby aims to
position Instagram as a somewhat collaborative element to its users, aiming at no more than
the most positive experience while using the app, and with no other motive to employ ranking
and recommender algorithms than to help discover new relevant content, staying up to date
with friends and family and to deliver its best version of itself. The way of overly positive
illustration could almost make the reader forget for a moment, that Instagram pursues own
economic interests, instead of being there to simply make its users' lives better.

Reality, however, tells us not only, that Instagram, parented by its owning company Meta
Platforms, generates astronomical revenue, but also, that algorithms employed all across
social media platforms these days, including Instagram, are indeed not users' best friends and
serve the primary purpose of sustaining user attention, maximizing their engagement and
doing their utmost to keep users on the app or website for as long as possible. The reason is
simple: more time spent on the app means more targeted advertising displayed to each
individual, which can in turn be directly translated into revenue for the platform.

ZeMV e-Publikation 05/2024 13 https://zemv.org

Zentrum für
Medienpsychologie und Gross, Michaud, Zerrouki, & Hamood
Verhaltensforschung

Even through mere logic alone, one must already fall into doubt about the true intentions
behind these algorithms. The notion, that a company of Instagram's magnitude would employ
the most sophisticated sort of contemporary technology not for their own economic benefit,
but rather for enhancing users' browsing experience is more than unrealistic; and even more
so, raises the question as to how it generates financial revenue that market analysts expect to
be north of USD 70 Billion in 2024 (WARC, 2023).

Investigations could reveal, that engagement mechanisms and algorithmic content curation in
Instagram are strategically designed to maximize user engagement and sustain attention,
which is the direct translation into economic benefit for the platform. Mosseri's assertion that
Instagram's algorithms aim to personalize user experience contradicts findings that highlight
the platform's focus on engagement metrics. For instance, studies show that likes and
comments significantly influence content visibility (Purba & Yulia, 2021). This reveals a
strategy centered around maximizing user interaction rather than merely enhancing user
satisfaction. The more engagement users exhibit on the platform, the more advertising
revenue can eventually be generated. Engagement metrics, such as comments, likes and
shares play a critical role in the platform's use of predictive algorithms and enhancement of
content visibility and user interaction. Through comprehensive extraction of distinct features,
the investigation could support the notion, that Instagram's algorithmic architecture inherently
aims to maximize and sustain attention and engagement (Tricomi et al., 2023). Especially in
hindsight of the estimated USD 33.25 billion in 2022 (ibid.), it is difficult to rationalize away
the economic motivation behind the algorithmic engagement-driven architecture. Contrary to
Mosseri's statement that the algorithms are designed to help users make the most of their time,
research clearly indicates that factors like image quality and posting time are optimized to
sustain user attention (Wang et al., 2020). In turn, it can be assumed that the primary goal is to
keep users engaged on the platform as long as possible, aligning with economic benefits for
Instagram.

It could further be scientifically demonstrated, how deep learning models can be utilized to
predict and enhance engagement on Instagram. By training personalized engagement
prediction models on individual accounts, Wang et al. (2020) were able to show the potential
for algorithms to sustain user engagement and attention through specific, highly engaging
content. Their examination indicates, that the underlying strategy in Instagram's ecosystem
supports a sophisticated data-driven approach to maximizing attention and prolonging
engagement, aligning with and reflective of broad industry trends, rather than proprietary
algorithm disclosure, as Mosseri's assertions might suggest. This research confirms the notion,
that the economic incentives behind high engagement make for better visibility on the
platform, that is directly tied to engagement-driven algorithms, which challenges Mosseri's
portrayal of algorithms as tools for enhancing user experience.

In diffusion modeling, Purba et al. (2021) could show, that engagement is the most critical
metric for realistic maximization of influence, suggesting, that Instagram's inherent
architecture relies heavily on high user interaction. Through introduction of the Engagement
ZeMV e-Publikation 05/2024 14 https://zemv.org
Zentrum für
Medienpsychologie und Gross, Michaud, Zerrouki, & Hamood
Verhaltensforschung

Grade (EG) and their diffusion models IC-eg and LT-eg, engagement metrics were
incorporated to demonstrate, how engagement is the essential component of how content is
propagated largely across Instagram, directly contradicting the portrayal of algorithmic
curation for an improvement of individual user experience.

Contrary to Adam Mosseri's statements

"We want to make the most of people's time, and we believe that using technology to
personalize everyone's experience is the best way to do that" (Mosseri, 2023)

and

"Our intention is to help creators reach their audiences and get discovered so they can
continue to grow and thrive on Instagram" (Mosseri, 2023),

research on the mechanisms and algorithmic architecture clearly shows, that the design is
built to surface content that maximizes engagement metrics, which can be seen as a strategy
to keep users engaged for longer periods, rather than a mere focus on helping creators;
likewise, rather than personalizing content curation for a better experience, the economic
structure of Instagram, its practices and its sources of revenue clearly reveal, that
prioritization of engagement metrics, such as likes, comments and shares foster frequent
interaction, longer durations spent on the app and thereby directly align with Instagram's
economic motives.

Findings from a 2021 study, conducted by Purba et al., substantiate these perspectives,
because it could be found, that various features can be extracted to predict Engagement Rates
and amplify combinations of favorable metrics to sustain user attention and increase
engagement. This evidence further contradicts the image of algorithms as a user-friendly tool
to discover more relevant content and supports the perspective of maximizing profitability
first and foremost.

In this paper, we would like to address another, deeper layer of concern about how the
statements made in the publication are vastly inaccurate, in the context of how users' behavior
is being tracked and analyzed; the effort, precision and shocking accuracy with which
algorithms can nowadays make estimations, classifications and predictions, is a topic that we
are certainly not the first to mention. However, deliberately conveying the impression, that

there is merely some data collected for the purpose of getting to know the interests of
the user better, in order to personalize the experience

is so far from reality, that in the context of debunking the publicly shared notion of
harmlessness and goodwill, they must be rectified, and to a large extent.

ZeMV e-Publikation 05/2024 15 https://zemv.org

Zentrum für
Medienpsychologie und Gross, Michaud, Zerrouki, & Hamood
Verhaltensforschung

Shoshana Zuboff wrote in her 2019 seminal work The Age of Surveillance Capitalism: The
Fight for a Human Future at the New Frontier of Power, that at the time of the investigation
and creation of her work, the depth and amount of extraction of individual behavioral data had
been unprecedented in history, going as deep as classifying individual phrases and patterns of
speech, which she labels the behavioral surplus, the meta-data or mid-level metrics, out of
which classifications and predictions can be made. She clarifies, that while seemingly banal,
through the amount and sophistication of data collection and analyses, including
contemporary methodology, there is a depth in behavioral insights, that is unimaginable.

Citing big data scientist Michal Kosinski, she writes

"As Kosinski told an interviewer in 2015, few people understand that companies such
as Facebook, Snapchat, Microsoft, Google and others have access to data that
scientists would never be able to collect." (Zuboff, 2019, p. 261; Kosinski, 2015)

and further

"In his 2015 interview, Kosinski observed that 'all of our interactions are being
mediated through digital products and services which basically means that everything
is being recorded.' He even characterized his own work as 'pretty creepy': 'I actually
want to stress that I think that many of the things that... one can do should certainly
not be done by corporations of governments without users' consent.'" (ibid.).

Delving deeper in Kosinski's research, accompanying this 2015 interview, we would like to
draw attention to a 2015 study, that is concerned with entirely computer-based judgments on
human personality, that are exactly based on such algorithmically extracted traits. They were
solely based on Facebook-Likes in the investigation, yielding the following results:

After a number of x analyzed Facebook likes only, the computer did a significantly better and
more accurate personality estimation and prognosis, than:

- a work colleague (x = 10)

- a friend (x = 70)
- a sibling or the subject's parents (x = 150)
- the subject's own spouse (x = 300).

While this (Youyou et al., 2015) serves as equally impressive and concerning evidence for the
capability of behavioral prediction and the exploitation of gathered data in social media, we
want to highlight, that this study does not serve as irrefutable evidence for Facebook's
practices, and also not as evidence for how Instagram leverages user behavioral data.

ZeMV e-Publikation 05/2024 16 https://zemv.org

Zentrum für
Medienpsychologie und Gross, Michaud, Zerrouki, & Hamood
Verhaltensforschung

However, it shows the vast potential of what can be done with such data, especially since the
data-trail left behind on social media does not only consist of likes, but much more.
Considering the accuracy that could be obtained from likes alone, it is fair to say, that through
algorithms which continually monitor user behavior on a multi-scale basis, the potential in
sophistication of behavioral prediction and manipulation can indeed be described as creepy.

Those insights are solidified by contemporary studies, that could reveal, that digital records
about patterns of Facebook usage, or even as little as the language used in social media posts,
are enough of a digital footprint to enable algorithms to extract highly detailed and accurate
personality traits to build sophisticated personality profiles from them, upon which behavior
can be precisely predicted. While users are unaware of those facts, the personal attributes that
could be derived included sexual orientation, ethnicity, political and religious views, general
personality traits, intelligence, overall hapiness, habits of substance abuse, parental separation,
age and gender (Kosinski et al., 2013; Bachrach et al., 2012; Park et al., 2015; Zuboff, 2019).

Adam Mosseri himself makes several assertions on the depth of monitoring, and while
making an effort to present their purpose as a benefit to Instagram's users, considering the
former paragraph, we may cite some of there assertions with a healthy portion of doubt:

"Likes and comments are important signals for ranking content in Feed and Stories."

"We look at how often you interact with the person who posted, such as liking or
commenting on their posts."

"We try to predict how likely you are to be interested in a photo or video based on past
behavior."

"How long you spend looking at a post also helps us understand what is interesting to
you."

"We consider your activity in Explore and Search to understand what you might be
interested in."

"If you visit someone's profile or send them a message, it's a strong signal that you're
interested in that person."

"We look at how often you share content, both publicly and privately."

ZeMV e-Publikation 05/2024 17 https://zemv.org

Zentrum für
Medienpsychologie und Gross, Michaud, Zerrouki, & Hamood
Verhaltensforschung

While these admissions provide an overview of the monitoring depth implemented by

Instagram, it is interesting to consider, that especially through the fourth statement, we can
derive that there is real-time surveillance, extracting behavioral data as soon as the app is
opened, which makes it obvious, that there is a continuous data-stream with every ever-so
slight move, stored by Instagram and fed into the algorithm. If we consider the sophistication
in behavioral predictions that could be derived from likes alone, and ten years ago, with less
sophisticated information technology, then at this point, it must become obvious, how
granular and accurate predictions and data processing potentially is today – and this still all
from the perspective of the obvious and official, irrespective at this point of the potentially
hidden data, methods and practices.

According to J. P. Titlow (2017), what makes Instagram decide about which content to show,
out of the millions of options, is behavioral metadata, and irrespective of the content. That
means, that the content of images is relatively unimportant for an algorithm in the decision
how to rank the content. Behavioral metadata includes much more than, as Titlow puts it, "if
you like this, you'll like that" logic: there is a highly complex interweb at play, well-hidden in
the background, that extracts endless datapoints, in order to increase its users' overall
engagement. According to Titlow, who argues Instagram "is mining the multilayered social
web between users", this strategy worked out brilliantly: considering the inauguration of
algorithmic personalization in the Explore feature of the app, two years after Facebook had
acquired Instagram, experiments saw a surge of 400 % in engagement.

Considering the economic side of that medal, data extracted from SignHouse, Business of
Apps and Statista suggest, that while Instagram was acquired by Facebook in 2012 for USD 1
billion, by late 2016 this skyrocketing engagement had already catapulted its valuation to
USD 30 billion. Personalized algorithms fall well into that timeframe.

There is significant proof, that engagement maximization is at the heart of Instagram's

business model, and certainly, sophisticated algorithms help well in extrapolating metrics
obtained from user behavior. Mosseri's denial of content suppression to boost ad sales is
contradicted by the platform's reliance on engagement metrics, which directly correlate with
advertising revenue. Research solidifies that the economic structure of Instagram heavily
relies on maximizing user engagement to drive profitability, challenging the notion that the
algorithms are solely for user benefit, as we will show in the following paragraph.

ZeMV e-Publikation 05/2024 18 https://zemv.org

Zentrum für
Medienpsychologie und Gross, Michaud, Zerrouki, & Hamood
Verhaltensforschung

Since parent company Meta, formerly Facebook, has an interest to extract utmost profits from
its acquisition, the combined data-streams of Instagram and Facebook (given an individual
actually uses both) would make for the ultimate resource of behavioral metadata (pun not
intended) to create ever-so sophisticated profiles; remembering the need of no more than 300
Facebook likes to actually psychologically profile a person more accurately than their own
spouse.

Reports of certain incidents let us safely assume, that this is exactly what happens, for
instance by the account of T. Stenovec (2016), who argues, that Facebook and Instagram share
user data among each other, to improve their personalization algorithms. According to
Stenovec, this has been confirmed to Tech Insider by an Instagram spokesperson.

In her book, Shoshana Zuboff (2019) describes the complexity that lies in the data building
the surplus in the context of the aftermath of the Cambridge Analytica scandal in 2018 –
which in our context can be seen as one strong argument against the oversimplified and
skewed portrayal of Instagram's algorithmic practices made by Mosseri (2023): as Zuboff
explains, there is a vast asymmetry of knowledge between the intelligence Facebook gathers
on its users and the knowledge thereof, obtained by its users themselves. Facebook argued, in
the wake of criticism, that letting users freely access exactly this intelligence would "require it
to surmount 'huge technical challenges'" (p. 453); furthermore, she states, that the data
Facebook would provide to users did not include the data on the behavioral surplus gathered
for the purpose of prediction products, eventually sold and employed for behavioral
modification.

Instagram's algorithms are mechanisms that exert considerable control over the sort of content
that is shown to users and in general, what content gains visibility, which highly benefits
Instagram's commercial interests. Instead of solely enhancing the experience for users, the
strategic interaction of actively shaping user behavior and maximizing visibility and
engagement illustrates complex power dynamics and how maintenance of high engagement
levels benefits the platform's profitability. The manipulation of content distribution on an
individual scale has therefore been deemed a visibility game (Cotter, 2018).

Those findings align with the knowledge from investigations of Instagram's recommendation
algorithms, that surfaced a significant influence of commercial optimization strategies, which,
according to Mehlhose et al. (2021), did not differ significantly from those in other social
networks: the clear priority is on engagement metrics that primarily drive commercial
benefits. Jaakonmäki et al. (2017) came to a similar conclusion, when they established, that
the algorithms embedded in Instagram, and generally in social media and its marketing
ecosystems, are primarily geared towards the outcome of maximizing engagement and user
attention. The implemented machine learning models prioritize features in their analyses of
content, that amplify interaction and consequently, profitability for the platform.

ZeMV e-Publikation 05/2024 19 https://zemv.org

Zentrum für
Medienpsychologie und Gross, Michaud, Zerrouki, & Hamood
Verhaltensforschung

Conclusive Statement

After careful review of the existing literature, Adam Mosseri's portrayal of Instagram's
algorithmic practices, and especially their intentions, hold little to no credibility. While
Mosseri publicly emphasizes user experience and creator support as motivation, evidence
clearly contradicts those claims. It could be found, that Instagram's algorithms are
meticulously designed to maximize user engagement and sustain their attention,
predominantly to the content that algorithms curate for them. The latter is meanwhile far from
chance or coincidence, instead it is carefully selected as a means to maximize profits. Factors
such as likes, comments, image quality, and user history are permanently monitored in real-
time and are strategically optimized to sustain prolonged interaction, well aligning with the
platform's economic interests. In addition, as we demonstrated, the infrastructural demands in
obtaining, retaining and processing the immense amounts of multilayered datapoints in a
sheer astronomical complexity have not only the potential to make extremely, even
worryingly accurate individual profile predictions and behavioral manipulations, they do also
have to pay off reasonably to justify the tremendous effort. "Enhancing the experience" is
therefore certainly not the primary motivation. Conclusions from the general advertising and
social media marketing ecosystem confirm those perspectives, especially since it could be
established, that the algorithmic architecture of Instagram's proprietary mechanisms does not
significantly differ from others. These findings debunk the notion of a purely user-centric
algorithmic design, revealing the underlying profit-driven motives and likewise, the deceptive
nature of Mosseri's publication.

ZeMV e-Publikation 05/2024 20 https://zemv.org

Zentrum für
Medienpsychologie und Gross, Michaud, Zerrouki, & Hamood
Verhaltensforschung

General Conclusion

As our research has been able to show, there is more to using algorithms than simply the best
experience for the user. Our analysis reveals significant discrepancies between the image that
Adam Mosseri likes to convey about Instagram's algorithmic practices and the realities. Both
lie far from each other.

Our sentiment analysis demonstrates a deliberate use of positive language and framing
techniques to create a user-friendly image of Instagram as a business, in attempting to
intentionally steer the reader's attention away from the fact, that Instagram's – as any
company's – primary intention is making a profit. While presenting Instagram as a tool, that
tries its utmost to align with each individual user's goals, it is especially critical for Instagram
to convey such a positive portrayal at a time where there is rising criticism of data mining
practices, general suspicion of algorithmic content curation, data privacy concerns and the
overall harmful practices exhibited by social media enterprises.

Understandably, there is a need for Instagram to maintain a good reputation, leading to

potentially deceptive narratives that are meant to obscure the economic motives behind
employing recommender algorithms. Therefore, user benefits are numerously emphasized
throughout the statements, leading to a strongly positive sentiment across the publications.
Yet, in the context of increasing scrutiny over data privacy and transparency claims, Adam
Mosseri's statements primarily serve the purpose of soothing public concerns and to deflect
criticism. Portraying algorithms as users' "friendly tool for a seamless and entertaining
experience" however, skews the public image of what is really happening behind the curtain.

As our detailed examination of existing literature could show, data extraction and processing
happens at a scale, that entirely eliminates any doubt about the real intentions, which are
creating highly sophisticated profiles of each individual through real-time monitoring each
and every action and creating prediction-ready user profiles from hundreds of thousands of
datapoints, that are so sophisticated, that an algorithm perhaps knows the person behind those
datapoints better than any human ever could; the complexity in processing that such datasets
require simply exceeds the capacity of human interaction and cognition, hence well-informed
scientists deemed their own findings creepy. Again, these practices, true not only for
Instagram and Facebook, but the entire industry of social media marketing, are far beyond
what Adam Mosseri attempts to convey. It can be safely said, that his publications are
understandable, from Instagram's perspective, however they are far from legitimate and are an
attempt to sugarcoat the use of algorithmic curation in the first place, while deliberately
concealing their inherent intention, vastly downplaying their true scope of manipulation and
malign potential.

ZeMV e-Publikation 05/2024 21 https://zemv.org

Zentrum für
Medienpsychologie und Gross, Michaud, Zerrouki, & Hamood
Verhaltensforschung

Heavy doubt can further be cast upon the portrayal of a flourishing environment for creators
and users around their close friends and family, positioning Instagram as a societal benefit. As
we were able to expose, the entire publication was a mere means to present the platform in a
more favorable light amid a general surge in criticism of the industry.

Two things however are certain; neither has Mosseri's statement shed any light on how
Instagram works – on the contrary – nor has anything about the industry's practices changed
until the present day.

5 Conflict of Interest Statement

As part of our commitment to diligence, academic integrity and transparency, we declare, that
none of the authors of this publication have any economic involvement in social media
networks or any related activities, and neither does the publisher of this work, the Zentrum für
Medienpsychologie und Verhaltensforschung (ZeMV).

While there is no conflict of interest in traditional sense, we consider it important to fortify

our credibility and to uphold academic rigor expected in our field. Therefore, while we
emphasize, that our research adheres to rigorous academic standards and is impartial and
objective, we highlight, that ZeMV is an entity that critically examines the impact of social
media in general. This includes Instagram as well as any other social media platform.

In order to ensure our audience is fully informed of our institutional perspective, we hereby
confirm that findings and conclusions presented in this publication are based on empirical
data and thorough scholarly analyses and are devoid of personal biases, external influences
and economic interests.

ZeMV, the publisher of this work, is a not-for-profit scientific entity, hosting scholars that
conduct research in media psychology and behavioral science.

ZeMV e-Publikation 05/2024 22 https://zemv.org

Zentrum für
Medienpsychologie und Gross, Michaud, Zerrouki, & Hamood
Verhaltensforschung

References
Bachrach, Y., Kohli, P., Kosinski, M., Stillwell, D., Graepel, T. (2012). Personality and Patterns of Facebook
Usage. Microsoft Research. Available at: https://www.microsoft.com/en-us/research/wp-content/uploads/
2016/02/FacebookPersonality_michal_29_04_12.pdf
Bellavista, P., Foschini, L., & Ghiselli, N. (2019). Analysis of Growth Strategies in Social Media: The
Instagram Use Case. 2019 IEEE 24th International Workshop on Computer Aided Modeling and Design of
Communication Links and Networks (CAMAD).

CaPPr. (2015). Interview with Michal Kosinski on Personality and Facebook Likes, May 20, 2015. Retrieved
from: https://www.youtube.com/watch?v=pJGuWKqwYRk

Cotter, K. (2018). Playing the visibility game: How digital influencers and algorithms negotiate influence on
Instagram. New Media & Society, 21(4), 895-913.

Daniel, C. Instagram Revenue and Growth Statistics (2024). (2023). SignHouse. Retrieved from: https://
www.usesignhouse.com/blog/instagram-stats

Dixon, S. J. (2014). Instagram - statistics & facts. Statista. Retrieved from: https://www.statista.com/topics/
1882/instagram/#topicOverview

Iqbal, M. (2024). Instagram Revenue and Usage Statistics. Business of Apps. Retrieved from: https://
www.businessofapps.com/data/instagram-statistics/

Jaakonmäki, R., Müller, O., & Brocke, J. (2017). The impact of content, context, and creator on user
engagement in social media marketing. HICSS 2017 Proceedings, 1-9.

Kosinski, M., Stillwell, D., and Graepel, T. (2013). Private Traits and Attributes Are Predictable from Digital
Records of Human Behavior. Proceedings of the National Academy of Sciences of the United States of
America 110,15, 5802–5.
Mehlhose, F. M., Petrifke, M., & Lindemann, C. (2021). Evaluation of graph-based algorithms for guessing
user recommendations of the social network Instagram. 2021 IEEE 15th International Conference on
Semantic Computing (ICSC), 409-414.

Mosseri, A. (2021). Shedding More Light on How Instagram Works. Retrieved from: https://
about.instagram.com/blog/announcements/shedding-more-light-on-how-instagram-works

Mosseri, A. (2023). Instagram Ranking Explained. Retrieved from: https://about.instagram.com/blog/

announcements/instagram-ranking-explained

Oliveira, L. M., & Goussevskaia, O. (2020). Sponsored content and user engagement dynamics on Instagram.
Proceedings of the 35th Annual ACM Symposium on Applied Computing, 124-131.

Park, G., Schwartz, H. A., Eichstaedt, J. C., Kern, M. L., Kosinski, M., Stillwell, D., Ungar, L. H., Seligman,
M. (2015). Automatic Personality Assessment Through Social Media Language. Journal of Personality and
Social Psychology 108 (6), 934–52.

Purba, K. R., & Yulia, Y. (2021). Realistic influence maximization based on followers score and engagement
grade on Instagram. Bulletin of Electrical Engineering and Informatics, 10, 1046-1053.

Purba, K. R., Asirvatham, D., & Murugesan, R. (2020). Classification of Instagram fake users using
supervised machine learning algorithms. International Journal of Electrical and Computer Engineering
(IJECE), 10(3), 2763-2772.

ZeMV e-Publikation 05/2024 23 https://zemv.org

Zentrum für
Medienpsychologie und Gross, Michaud, Zerrouki, & Hamood
Verhaltensforschung
Purba, K. R., Asirvatham, D., & Murugesan, R. (2021). Instagram post popularity trend analysis and
prediction. Proceedings of the International Conference on Information Management and Technology
(ICIMTech 2020).
Purba, K. R., Asirvatham, D., & Murugesan, R. (2022). Influence maximization and diffusion models based
on engagement and activeness on Instagram. Journal of King Saud University - Computer and Information
Sciences, 34 (6), 2831-2839.

Skrubbeltrang, M. M., Grunnet, J., & Tarp, N. T. (2017). #RIPINSTAGRAM: Examining user's counter-
narratives opposing the introduction of algorithmic personalization on Instagram. First Monday, 22(4).

Stenovec, T. (2016). How to stop Instagram ads from following you. Business Insider. Retrieved from:
https://www.businessinsider.com/how-to-stop-instagram-ads-from-following-you-2016-3

Titlow, J. P. (2017). How Instagram Learns From Your Likes To Keep You Hooked. Fast Company. Retrieved
from: https://www.fastcompany.com/40434598/how-instagram-learns-from-your-likes-to-keep-you-hooked

Tricomi, P. P., Chilese, M., Conti, M., & Sadeghi, A. (2023). Follow us and become famous! Insights and
guidelines from Instagram engagement mechanisms. Proceedings of the 15th ACM Web Science Conference
2023.
Wang, L., Liu, R., & Vosoughi, S. (2020). Salienteye: Maximizing engagement while maintaining artistic
style on Instagram using deep neural networks. Proceedings of the 2020 International Conference on
Multimedia Retrieval.

WARC. (2023). Instagram forecast to hit $71bn revenue by 2024. Retrieved from: https://www.warc.com/
content/feed/instagram-forecast-to-hit-71bn-revenue-by-2024/en-GB/8650

Wirtschaftspsychologie Aktuell. (2018). Facebook kennt dich besser als deine Freunde. Retrieved from:
https://wirtschaftspsychologie-aktuell.de/magazin/leben/facebook-kennt-dich-besser-als-deine-freunde

Youyou, W., Kosinski, M., & Stillwell, D. (2015). Computer-based personality judgments are more accurate
than those made by humans. Proceedings of the National Academy of Sciences, 112(4), 1036-1040.

Zou, L., Xia, L., Ding, Z., Song, J., Liu, W., & Yin, D. (2019). Reinforcement learning to optimize long-term
user engagement in recommender systems. Proceedings of the 25th ACM SIGKDD International Conference
on Knowledge Discovery & Data Mining.

Zuboff, S. (2019). The age of surveillance capitalism: The fight for a human future at the new frontier of
power. Public Affairs.

ZeMV e-Publikation 05/2024 24 https://zemv.org

 Zentrum für
Medienpsychologie und Gross, Michaud, Zerrouki, & Hamood
Verhaltensforschung

Annex: Code Disclosure (Segment n = placeholder)

ZeMV e-Publikation 05/2024 25 https://zemv.org

View publication stats