=Paper=
{{Paper
|id=Vol-3639/paper5
|storemode=property
|title=Refining Deliberative Standards for Online Political Communication: Introducing a
Summative Approach to Designing Deliberative Recommender Systems
|pdfUrl=https://ceur-ws.org/Vol-3639/paper5.pdf
|volume=Vol-3639
|authors=Sjoerd B. Stolwijk,Corinna Oschatz,Michael Heseltine,Damian Trilling
|dblpUrl=https://dblp.org/rec/conf/normalize/StolwijkOHT23
}}
==Refining Deliberative Standards for Online Political Communication: Introducing a
Summative Approach to Designing Deliberative Recommender Systems
==
https://ceur-ws.org/Vol-3639/paper5.pdf
Refining Deliberative Standards for Online Political
Communication: Introducing a Summative Approach
to Designing Deliberative Recommender Systems
Sjoerd B. Stolwijk1 , Corinna Oschatz1 , Michael Heseltine1 and Damian Trilling1
1
Amsterdam School of Communication Research (ASCoR), University of Amsterdam, Netherlands
Abstract
Measuring deliberative debate quality is an emerging topic in computational work, since it allows
applying deliberative democratic ideas in an online domain. When doing so, many studies follow
Habermas [1, 2, 3] in defining norms of online deliberative debate quality. They then proceed to propose
and test new ways to measure indicators like equality, diversity, interactiveness, rationality, and civility.
Consequently, implementing them within recommender systems is the necessary next step to realize
those values in online communication. Although this is important work, we argue that recent advances in
political science suggest that constructing a system which produces such a deliberative debate is unlikely
to, by itself, contribute in an optimal way to deliberative democracy at a societal scale. Instead, we
propose a complementary, summative approach to designing deliberative recommender systems. It treats
online platforms as complementary to other communication channels, and argues for optimizing how
to best facilitate (summative) deliberation at a societal scale rather than perfecting (micro) discussions
between citizens. We illustrate this with an example of how a news recommender system based on a
summative approach would have to be designed vis-a-vis a more traditional, additive approach.
Keywords
deliberative democracy, normative standards, online debate quality, computational text analysis, moral
recommender systems
1. Introduction
The scale and growing influence of online communication holds great promise to realize
deliberative democratic ideals. Normative theory provides an alternative to revenue-based
models to develop recommender systems that seek more positive effects for society.1 But while
many online platforms offer the potential for users to interact with their content as well as
with other users and thereby create the potential for political discussion, it is far from clear
how recommender systems need to be designed to achieve positive contributions to political
NORMalize 2023: The First Workshop on the Normative Design and Evaluation of Recommender Systems, September 19,
2023, co-located with the ACM Conference on Recommender Systems 2023 (RecSys 2023), Singapore
$ s.b.stolwijk@uva.nl (S. B. Stolwijk); c.m.oschatz@uva.nl (C. Oschatz); m.j.heseltine@uva.nl (M. Heseltine);
d.c.trilling@uva.nl (D. Trilling)
https://www.uva.nl/profiel/s/t/s.b.stolwijk/s.b.stolwijk.html (S. B. Stolwijk); https://damiantrilling.net
(D. Trilling)
0000-0003-1421-2445 (S. B. Stolwijk); 0000-0003-3276-5721 (C. Oschatz); 0000-0002-2943-4414 (M. Heseltine);
0000-0002-2586-0352 (D. Trilling)
© 2023 Copyright for this paper by its authors. Use permitted under Creative Commons License Attribution 4.0 International (CC BY 4.0).
CEUR
Workshop
CEUR Workshop Proceedings (CEUR-WS.org)
Proceedings
http://ceur-ws.org
ISSN 1613-0073
1
See Knudsen [4] for an illustration of how recommender systems might influence democratic debate values
1
CEUR
ceur-ws.org
Workshop ISSN 1613-0073
Proceedings
Sjoerd B. Stolwijk et al. CEUR Workshop Proceedings 1–12
discourse at a societal scale. On the one hand, the development online deliberative debate
indicators is a flourishing field [5] – which is very good, as after all, recommender systems
need some metrics to optimize for or be evaluated on. On the other hand, from a normative
point of view, these metrics also need to be grounded in democratic theory – and this is less
straightforward than it my seem.
To begin with, scholars who wish to develop recommender systems that facilitate online
discussions along normative democratic lines, have to choose which out of various alternative
conceptions of democracy they want to rely on [6, 7]. For example, based on so-called liberal
theories of democracy one may argue that topics that are prominent on the political agenda
should also be prominent in a recommendation, while based on so-called critical theories,
one would expect marginal topics to be recommended instead; so-called participative models
would draw attention to topics citizens “should know”, and so-called deliberative models would
focus more on the diversity of the items [6]. In this paper, we focus on deliberative models of
democracy. Deliberative democracy is often considered the most demanding in terms of the
required quality of (online) political discussion, and thereby provides a nice ideal point to strive
for (cf. [8]). Its emphasis on discussion also fits well with the nature of online communication
platforms.
Although there are many variants of deliberative theory, the work of Habermas [1, 2, 3] is
considered a cornerstone in this literature (c.f.[9, 5]). Deliberative democracy is about more than
making decisions based on aggregating the preferences of citizens, or even than preferences
based on accurate information, but rather holds that (collective) preferences should be formed
through inclusive, reasoned debate [10, 8]. In this view, online communication needs to provide
relevant and accurate information, but also connect citizens and motivate them to share and
debate their views in a deliberative way.
Scholars seeking to measure deliberation online have proposed and listed various indicators
which can be used by recommender systems for this purpose. They measure concepts like
equality, diversity, rationality, interactivity, civility and reference to the common good [7, 5,
11, 12]. Recommender systems are widely used on for example E-commerce platforms, social
networks, video-sharing platforms, and news websites and apps. Each of these platforms offer
the potential for politically relevant discussions [13]. However, most computational work
focuses on developing specific indicators relevant to deliberative democracy (e.g. [14], but
recommender systems focusing on all indicators are needed to fully realize deliberative ideals
[5]. In addition, less attention has been paid to whether and how these indicators can best be
put to use to realize deliberative democratic ideals on a societal scale.
In political science, scholars have struggled with similar issues when studying the deliberative
values of citizen participation initiatives. Although they are using non-computational methods,
their insights can be of use to the computational community as well. Similar to computational
scholars, deliberative political scientists started out evaluating whether and to what degree
the elements of deliberative democratic debate could be found in the exchanges between
citizens. After three decades of sustained, wide academic attention for this topic, they are now
moving away from what is called an “additive” approach to deliberation and question whether
deliberative quality is best treated as a single, unitary, concept consisting of all of its elements
to an equal degree regardless of context [15, 16]. The additive approach holds that "Deliberation
[...] is produced by specific methods or institutions which then add it—inject it, if you will—into
2
Sjoerd B. Stolwijk et al. CEUR Workshop Proceedings 1–12
the system more broadly" [15, p.7]. That is, they sought to construct an ideal forum to foster
deliberation in all its facets and use this as a way to produce deliberative democracy at a societal
scale. One example is forming a Citizen Assembly in which a stratified sample of the population
is invited to deliberate on a policy proposal over several days and then share their arguments
and propose their decision to the public at large and a legislative body in particular (e.g. see
[17]). Deliberation in this view is often seen in a more-is-better fashion, so more diversity in a
news feed is better for deliberative democracy than less diversity, more interactivity is better
than less and so on [16]. Although intuitive, the additive approach has been criticized for failing
to grasp the complexity of human behavior and for being ill-equipped to relate the outcomes of
individual (online) interactions back to the over-arcing goals of deliberative democracy on a
societal scale [18, 15]. Notably, even Habermas himself has recently questioned the approach
of using his indicators as a yardstick for societal debate, stating that “I do not see deliberative
politics as a farfetched ideal against which sordid reality must be measured” [19, p. 149].
In the next section, we outline the critique political scientists have levelled against the
additive approach. Subsequently, we sketch different views on how (micro) online deliberations
fit within the larger deliberative democratic ideal. Finally, we propose an alternative conception
of debate quality for online platforms, based on the recently proposed systematic, summative
approach to deliberative democracy [20, 15]. The summative approach does not seek to optimize
deliberation at any single venue, but rather maximize its value at the aggregate, societal scale.
This alternative treats online platforms as complementary rather than substitutive to traditional
media, and seeks to realize deliberative goals in a summative rather than additive fashion.
2. More Is Better, Or Is It?
Up until recently, either implicitly or explicitly, proponents of computationally measuring
deliberative quality often appear to advocate an additive, more-is-better approach. Some
describe how a specific debate aspect, like diversity, is helpful for deliberative democracy and
then propose how to measure it, others list indicators a good debate needs in order to be
deliberative [5, 12, 7, 6]. Like early deliberative political scientists, computational scholars
then evaluate whether a particular debate fulfills these criteria and to which degree (e.g. [21]).
A recommender system based on this approach could either aggregate indicator scores into
a total deliberative score to rank which items to recommend, or define minimum/maximum
values needed to pass platform moderation (such as for civility). However, doing so would
imply assuming that deliberation is a unitary concept, i.e. that all criteria need to be fulfilled
together to a certain degree for the ascribed benefits to deliberative democracy to materialize.
While intuitive, Mutz [18] argues that one should be careful, since this approach carries some
assumptions about human behavior that are unwarranted. Also adding the scores of various
criteria together to one overall deliberative score implies that deficiencies on the score for one
criteria can be compensated for by higher scores on other criteria, while using indicators for
minimum benchmark values implies that failing to meet the minimum level on one criteria
disqualifies any achievements on other criteria.
A growing literature questions that deliberation is a unitary concept and all indicators nec-
essarily need to go together (e.g. [18, 15]). In fact, there are many situations where various
3
Sjoerd B. Stolwijk et al. CEUR Workshop Proceedings 1–12
indicators of deliberation appear to be at odds with one another [16]. For example, diversity
might conflict with inclusion: the more diverse the voices in a debate, the less likely it is that
everyone participates, since debate requires conflicting arguments and most people feel uncom-
fortable being confronted by (too many) opposite opinions (cf [22, 23]). Likewise representation,
accountability and openness might conflict with civility: publicity makes representatives more
accountable to their supporters, but is also found to be less conducive to mutual respect and
constructive politics than deliberation behind closed doors [15]. This directly highlights one
of the main problems in deliberative democratic theory: that quality aspects of debates are
frequently conflated with positive effects of deliberative debates in definitions of what counts as
deliberative democracy [18, 24]. This lack of conceptual clarity in stipulating causes and effects
masks the myriad in conceptualizations of what exactly constitutes deliberation as well as what
its effects are supposed to be.
Like deliberative indicators, deliberative effects/outcomes are also more complex empirically
than a unitary concept approach would appear to imply. A great many presumed effects can be
found in the literature. Bächtiger and Parkinson [15] group them as: (1) epistemic outcomes
(find best possible approach to handle a common problem); (2) ethical outcomes (follow the
rational argument dominated deliberative process as a goal in itself); (3) providing legitimacy
(of collective decisions formed through deliberation); (4) emancipation of minority groups
(providing a space to make all citizens heard); (5) transformation and clarification of preferences
(people learn from the debate and change their views or deepen their perspectives in return);
with some also listing (6) consensus as a desired outcome. Although these goals overlap to a
certain degree, different deliberative aspects contribute to them in a different degree. Scholars
have listed many contradictions between different aspects of deliberation and deliberative
outcomes. For example, if the process of rational argumentation is the goal, the resulting
formal-tone of the debate might be off-putting to some citizens, and dominant groups might
use their definition of what counts as a rational argument to suppress minority voices, also
requiring each position to come with an elaborate set of supporting arguments might favor
established and well documented positions over new voices that as of yet have not had the time
and space to develop such arguments [25, 26].
Given these conflicts between indicators of deliberation both among themselves and in relation
to the desired deliberative outcomes, Mutz [18] proposes to abandon the unitary concept of
deliberation and instead investigate which deliberative aspects have which effects under which
conditions. She argues that it is likely that different deliberative aspects might interact with
each other to reach specific outcomes, and should therefore be studied separately as well as in
different combinations. Applying this line of argumentation to recommender system design,
we can say that if a system optimizes either for (1) all deliberative indicators, or (2) an indexed
value of debate quality based on combining various indicators, or (3) some minimum/maximum
values of deliberative indicators, then the contradictions between these indicators and outcomes
as described above might lead this system to hinder rather than enable the deliberative process.
In the next section, we will elaborate more on the argument that what is seemingly good for
deliberation on a micro-level does not necessarily lead to good deliberative outcomes on the
societal macro-level.
4
Sjoerd B. Stolwijk et al. CEUR Workshop Proceedings 1–12
3. The Gap Between Micro and Macro
The second main critique of political scientists on using deliberative indicators as a yardstick to
measure contributions to normative standards of deliberative democracy, is the problematic
relation between specific debates (micro) and deliberative democracy on a societal scale (macro)
[27]. Most accounts of deliberative democracy aim at the societal level rather than that of the
individual debate. They require the deliberative process is in fact democratic: that it culminates
in a collectively binding decision [16]. Habermas [19] stipulates how issues, information and
arguments in debates between citizens are picked up by societal actors, like social movements,
PR-organizations, political parties and the media which translate positions into coherent dis-
courses relating distinct positions to relevant arguments, which then feed into the political
arena to result in collective decisions. He appears to be unsure, though, how to fit online debates
productively into his framework, and argues that such debates might actually be counterpro-
ductive warning that “[t]he platforms do not offer their emancipated users any substitute for
the professional selection and discursive examination of contents based on generally accepted
cognitive standards” [19, p. 160], and “the increasing dissonance of a strident diversity of
voices and the complexity of the challenging topics and positions is leading a growing minority
of media consumers to use digital platforms to retreat into shielded echo chambers of the
like-minded. For the digital platforms not only invite their users to spontaneously generate
intersubjectively confirmed worlds of their own but seem to lend the stubborn internal logic of
these islands of communication, in addition, the epistemic status of competing public spheres”
[19, p. 162]2 . On top of this, deliberation between all citizens in a modern society is practically
unfeasible due to constraints in time and resources, and can therefore only be realized at the
institutions of the state [19]. However, empirical research into the political arena, where the final
deliberative debate between contrasting discourses should culminate into collective decisions,
finds that parliamentary debate is oriented towards voting rather than aggregating information
and participants rarely change their preferences in view of contrasting information [15]. It is
therefore unclear whether and how online political communication contributes to deliberative
democracy at all.
This focus on relating the micro to the macro is often labelled as the “systemic turn”, as
it views deliberative democracy as a larger system rather than a debate at large [27, 29, 20].
Political scientists have proposed three main theoretical frameworks that specify how specific
debates between citizens could translate into collectively binding decisions in a deliberative
way. Bächtiger and Parkinson [15] group them into discursive, sequential and spatial models of
deliberation. The discursive model focuses on how people understand and shape society through
discourses that find their way into arguments, decisions and policy [30]. Various versions of
the sequential model hold that feedback loops ensure deliberative outcomes over time, where
societal debate influences political decisions, which are in turn part of societal debate to critique,
alter, maintain or reject them at another round of political decisions. The spatial model specifies
distinct deliberative functions of distinct institutions and the proper relations between these
institutions to ensure deliberative outcomes. Each of these models has received its fair portion
2
Note that Zuiderveen Borgesius et al. [28] found little empirical support for the relation between recommender
systems and echo chambers or filter bubbles
5
Sjoerd B. Stolwijk et al. CEUR Workshop Proceedings 1–12
of critique, where the discursive model is unclear how deliberation contributes in what way
to the forming of discourses and how these discourses feed into policy in a deliberative way,
the sequential model is often empirically incorrect in policy preceding debate especially for
non-salient issues, and the spatial model is found to be to static to encompass the creatively
changing nature of political decision making with new forms or organisation continuously
popping up (e.g. #MeToo) and actors reinterpreting their role and using their institution in new
ways [15].
4. Where Do Online Platforms Fit in the Larger System of
Deliberative Democracy?
Regardless of whether one adopts a discursive, sequential or spatial perspective, for an online
platform to contribute to deliberative democracy optimally it thus needs to produce some
deliberative contribution and transfer this contribution in some way to the wider society and
its political decision making bodies in particular. So what kind of contribution should online
platforms, via the recommender systems employed by them, make in the normative ideal of
deliberative democracy? Habermas [19] ascribes them a similar role to traditional media in his
(sequential) approach to deliberative democracy, and then criticizes online platforms for failing
to live up to those expectations: Platforms lack journalistic moderation and do not “qualitatively
filter opinions”, like the traditional media, where journalists scrutinize arguments and opinions
for facts and counterarguments and professionally select what to present to their audience
[19]. In addition, he is wary of personalization, since it may enable selective exposure and
echo chambers. On top of this online platforms have not been very successful empirically in
producing good deliberative debates. Where additive approaches to deliberative democracy
seek to find sites that facilitate optimal deliberation and then to channel the results as best as
possible to other parts of the deliberative system, most major online platforms are known to
lack those deliberative qualities [31, 32, 33].
However, there might be a different contribution that online platforms can make to delibera-
tive democracy, which is more feasible and better suited to their qualities. Habermas [19] notes
that the contribution of political communication in the public sphere, where online platforms
are located, is inherently limited, since only representative bodies make collective decisions.
The normative requirements of achieving deliberation in all its facets need thus not be so strict
for these platforms. As the discussion in Sections 2 and 3 has showed, it might not be optimal to
strive for deliberation in all its facets on online platforms to begin with. Bächtiger and Parkinson
[15] have recently proposed an alternative route to (macro) deliberative democracy which might
better fit with the potential of online platforms: the summative approach. They explain that
“the deliberative quality may emerge from the complex interactions of a variety of practices
and institutions rather than an input generated by one or two of them” [15, p. 14]. Simply put:
deliberative outcomes may be realized through non-perfect deliberative components, like online
platforms.
We propose that by thinking of deliberative democracy as a summative quality, we arrive at
other goals that recommender systems of online platforms should perform. They no longer need
to facilitate deliberation between citizens as best as possible, but might focus on optimizing the
6
Sjoerd B. Stolwijk et al. CEUR Workshop Proceedings 1–12
larger goals of deliberative democracy: reaching collective decisions on a rational basis involving
as many citizens on in the most equal way possible (cf. [16]). It needs some mechanism to (1)
involve citizens including those acting on behalf of social groups, politicians, PR officials etc.; to
(2) make them share their views and information; to (3) facilitate them to interact with each
other, existing discourses, and actions and words of political actors to develop and question
their opinion; to (4) collect and scrutinize arguments and positions into coherent discourses
concerning collective issues; to (5) communicate those discourses back to as many citizens as
possible, but (6) also to their representatives in the political arena. We believe a normative
design of recommender systems can play a role in each of these mechanisms.3
5. Moving Forward: A Complementary Role Within a
Summative Approach
We propose that in this summative understanding of deliberative democracy it is more helpful
to think of online communication as complimentary to other forms of political communication,
rather than as a substitute for traditional media. Our approach thus allocates a different function
to online platforms within the spatial and sequential system of deliberative democracy than
an additive approach would. For example, when facilitating debate between citizens, instead
of aiming for civil conversation, it might be better for (macro) deliberative democracy, if in
some cases people are allowed some incivility to make suppressed voices heard or to create a
communicative environment where some might feel more at home, where they feel they don’t
need to be eloquent and highly educated to be allowed to speak up. While at the same time,
those who might be put off by such discourse could be shielded from exposure to uncivil content
(cf. [34]). Where the one-to-many format of traditional media necessitates compromises in form
and content to fit a larger audience at the cost of individual differences, personalization enables
online platforms to tune into the individual needs of each citizen [22]. Hereby, content can
be presented to each citizen in a fashion tailored to encourage involvement, both in engaging
with arguments and in building the efficacy needed to share one’s views and information
(cf. [6]). Online platforms also provide opportunities to go beyond what can be achieved in
deliberative terms by traditional media, by directly linking citizens to, for example, journalists
and politicians [31]. Hereby they create a crucial link in facilitating deliberative sub-products,
like suppressed voices and new positions and arguments, to reach the traditional media and
institutional political arena.
Of course, such personalization is exactly what Habermas [19] criticises when he warns
about the the potential of creating parallel public spheres. When each citizen receives her
own tailor made version of online content, this might hinder a common understanding of the
main issues, positions and arguments facing a society. However, from the viewpoint of the
complementary role of online communication to other media and institutions, this can also
be seen as an opportunity for recommender systems to provide that common information
through sharing relevant content as provided by, for example, traditional media, politicians
or activist groups, in a tailored way to the largest audience. In this way online platforms can
3
Wessler [31] provides an alternative list of possible contributions of non-deliberative media to deliberative democ-
racy.
7
Sjoerd B. Stolwijk et al. CEUR Workshop Proceedings 1–12
actually improve the deliberative value of intersubjective understanding by involving previously
disaffected groups.
Table 1 provides an example of how this summative approach to designing deliberative
recommender systems differs from existing approaches in the parameters that need to be
optimized. The left column (“additive deliberative recommender system”) shows how additive
approaches seek to optimize all aspects of deliberation at once, and facilitate deliberative
democracy through concrete instances of citizen deliberation on online platforms, while the
right column shows that the summative approach instead optimizes deliberative outcomes at a
societal level. Note how the goals listed in the right column match the mechanisms required for
deliberative democracy outlined in Section 4. The right column focuses on optimizing exposure
to foster the deliberative value of inclusion; on optimizing engagement to get citizens to interact
with the debate; on optimizing the sharing of information to include as many insights from as
many citizens as possible; on providing these insights to other users and political actors alike;
and on including fact-check information to debunk misinformation and increase the factual
quality of the arguments. The summative column thus seeks to explicitly and directly link
citizens and political actors (e.g., politicians, but also including activist groups or PR-agencies),
since connecting the diversity of arguments leveled by both groups to each other is a specific
macro deliberative democratic value. While the additive approach (left column) thus seeks to fit
the debate into a deliberative mold, the summative approach (right column) seeks to optimize
societal deliberative outcomes.
The summative indicators proposed here are familiar ones in the field of recommender
systems and partly overlap with both additive indicators and those used in commercial revenue-
based applications. They are not meant to form a definitive list. They should rather be seen
as an invitation to scholars to propose their own more effective set of indicators to make
communication on online platforms contribute to the mechanisms required for deliberative
democracy outlined in Section 4. The overlap with existing commercial recommender systems
makes the summative approach more in line with existing practices on online platforms and
potentially easier to realize (cf. [35]). It does not try to change what people like about online
platforms, rather to guide them in a normative, societal deliberative direction. The familiarity
of these indicators illustrates the feasibility of this alternative route to realizing deliberative
values online.
6. Conclusion
Computational approaches to measuring deliberative indicators of online communication are a
blooming field and much work has been done in constructing indicators for various aspects
of deliberation, such as equality, rationality, interactivity, diversity and civility [5]. However,
political scientists have levelled two main critiques against the common computational imple-
mentation of debate quality: deliberation is unlikely to be related to deliberative democracy in
a unitary, more-is-better fashion and that (micro) online deliberation is unlikely to contribute
in an additive way to (macro) deliberative democracy. So even if computational scholars could
find a way to overcome the current technical challenges and construct a perfect set of reliable
and valid indicators of deliberative quality (see [5]), then still it would be questionable how
8
Sjoerd B. Stolwijk et al. CEUR Workshop Proceedings 1–12
Table 1
Examples of optimalization parameters for additive versus summative deliberative recommender systems
Additive deliberative recommender Summative deliberative recommender
system system
Typical metrics • equality • exposure
• diversity • engagement (likes/comments)
• rationality • sharing information
• interactivity • diversity of traditional news expo-
• civility sure
• diversity of user and political actor
exposure
• inclusion of fact-check info where
possible
Personalization Metrics matter for everyone equally Weight of metrics determined on in-
dividual basis
Temporal structure Static, all metrics are important at ev- Metrics can also be optimized for se-
ery point in time quentially; long-run outcome more
important than simultaneously good
scores on every metric
Contribution Realize deliberation within platform Contribute to societal deliberation
these indicators could be implemented in recommender systems to attain normative deliberative
outcomes.
We propose that one way out of this dilemma could be to build on the summative approach,
which seeks to optimize deliberative outcomes at a societal scale, rather than the additive
approach which seeks to optimize deliberation at each site/venue. Instead of a straightforward
design of recommender systems that either keep indicators of (micro) deliberation within
acceptable bounds or that optimize for them, we have argued that the complexity of human
behavior frustrates those efforts, and that it might lead to counterproductive results at a societal
level. Instead, we propose a summative approach to designing deliberative recommender
systems for online platforms. These systems take more account of the place of online platforms
within the larger system of deliberative democracy, and respect the potential trade-offs between
different deliberative values. They select and optimize an alternative set of indicators directed
at macro deliberative goals.
This approach aims to be both more fitting to the less than pure deliberative nature of online
debate (cf. [31, 32]. In fact, designing a summative deliberative recommender system does
not have to be at odds with commercial interests. For example, in the summative deliberative
framework, one explicit goal is to increase exposure to and engagement with the “long tail”
of content to make users aware of perspectives they may not be aware of, and allow them to
contribute. But this can perfectly align with commercial interests: In many recommendations
scenarios, it is an explicit goal to increase usage of long-tail items that the user would otherwise
not find.
Through its better fit with both deliberative democracy at the societal level and the nature of
9
Sjoerd B. Stolwijk et al. CEUR Workshop Proceedings 1–12
online platforms, the summative approach proposed here can help recommender systems to
increase the contribution that online communication can make to deliberative democracy and
thereby also help reduce negative effects often associated to online communication, such as
filter bubbles, selective exposure and misinformation [36, 37, 34].
Acknowledgments
This research is fully funded by the European Union and part of the “TWON: Twins of On-
line Networks” EU HORIZON grant, project number 101095095. We would like to thank the
anonymous reviewers whose detailed comments helped improve the paper. All opinions and
remaining errors are our own.
References
[1] J. Habermas, Strukturwandel der Öffentlichkeit. Untersuchungen zu einer Kategorie der
bürgerlichen Gesellschaft, Luchterhand, Neuwied and Berlin, 1962.
[2] J. Habermas, The theory of communicative action, Beacon Press, Boston, 1984.
[3] J. Habermas, Between Facts and Norms: Contributions to a Discourse Theory of Law and
Democracy, The MIT Press, Cambridge, MA, 1996.
[4] E. Knudsen, Modeling news recommender systems’ conditional effects on selective expo-
sure: evidence from two online experiments, Journal of Communication 73 (2023) 138–149.
doi:10.1093/joc/jqac047.
[5] A. Goddard, A. Gillespie, Textual Indicators of Deliberative Dialogue: A Systematic Review
of Methods for Studying the Quality of Online Dialogues, Social Science Computer Review
(2023) 089443932311566. doi:10.1177/08944393231156629.
[6] N. Helberger, On the Democratic Role of News Recommenders, Digital Journalism 7 (2019)
993–1012. doi:10.1080/21670811.2019.1623700.
[7] S. Vrijenhoek, M. Kaya, N. Metoui, J. Möller, D. Odijk, N. Helberger, Recommenders
with a Mission: Assessing Diversity in News Recommendations, in: Proceedings of the
2021 Conference on Human Information Interaction and Retrieval, ACM, Canberra ACT
Australia, 2021, pp. 173–183. doi:10.1145/3406522.3446019.
[8] J. Strömbäck, In Search of a Standard: four models of democracy and their norma-
tive implications for journalism, Journalism Studies 6 (2005) 331–345. doi:10.1080/
14616700500131950.
[9] D. Caluwaerts, K. Bernaerts, R. Kesberg, L. Smets, B. Spruyt, Deliberation and polarization:
a multi-disciplinary review, Frontiers in Political Science 5 (2023) 1127372. doi:10.3389/
fpos.2023.1127372.
[10] D. G. Freelon, Analyzing online political discussion using three models of demo-
cratic communication, New Media & Society 12 (2010) 1172–1190. doi:10.1177/
1461444809357927.
[11] M. Nelimarkka, J. P. Rancy, J. Grygiel, B. Semaan, (Re)Design to Mitigate Political Polar-
ization: Reflecting Habermas’ ideal communication space in the United States of America
10
Sjoerd B. Stolwijk et al. CEUR Workshop Proceedings 1–12
and Finland, Proceedings of the ACM on Human-Computer Interaction 3 (2019) 1–25.
doi:10.1145/3359243.
[12] L. Oswald, Automating the Analysis of Online Deliberation? A comparison of manual and
computational measures applied to climate change discussions, SocArXiv (2022).
[13] M. E. Wojcieszak, D. C. Mutz, Online Groups and Political Discourse: Do Online Discussion
Spaces Facilitate Exposure to Political Disagreement?, Journal of Communication 59 (2009)
40–56. doi:10.1111/j.1460-2466.2008.01403.x.
[14] S. Vrijenhoek, G. Bénédict, M. G. Granada, D. Odijk, M. de Rijke, RADio – Rank-Aware
Divergence Metrics to Measure Normative Diversity in News Recommendations, 2022.
URL: http://arxiv.org/abs/2209.13520, arXiv:2209.13520 [cs].
[15] A. Bächtiger, J. Parkinson, Mapping and Measuring Deliberation: Towards a New Deliber-
ative Quality, 1 ed., Oxford University Press, 2019. doi:10.1093/oso/9780199672196.
001.0001.
[16] D. F. Thompson, Deliberative Democratic Theory and Empirical Political Science, An-
nual Review of Political Science 11 (2008) 497–520. doi:10.1146/annurev.polisci.11.
081306.070555.
[17] K. Már, J. Gastil, Do Voters Trust Deliberative Minipublics? Examining the Origins and
Impact of Legitimacy Perceptions for the Citizens’ Initiative Review, Political Behavior
(2021). doi:10.1007/s11109-021-09742-6.
[18] D. C. Mutz, Is Deliberative Democracy a Falsifiable Theory?, Annual Review of Political
Science 11 (2008) 521–538. doi:10.1146/annurev.polisci.11.081306.070308.
[19] J. Habermas, Reflections and Hypotheses on a Further Structural Transformation of
the Political Public Sphere, Theory, Culture & Society 39 (2022) 145–171. doi:10.1177/
02632764221112341.
[20] J. Mansbridge, J. Bohman, S. Chambers, T. Christiano, A. Fung, J. Parkinson, D. F. Thomp-
son, M. E. Warren, A systemic approach to deliberative democracy, in: J. Parkinson,
J. Mansbridge (Eds.), Deliberative Systems, 1 ed., Cambridge University Press, 2012, pp.
1–26. doi:10.1017/CBO9781139178914.002.
[21] N. Beauchamp, Modeling and Measuring Deliberation Online, in: B. Foucault Welles,
S. González-Bailón (Eds.), The Oxford Handbook of Networked Communication, Oxford
University Press, 2020, pp. 320–349. doi:10.1093/oxfordhb/9780190460518.013.23.
[22] M. Reuver, N. Mattis, M. Sax, S. Verberne, N. Tintarev, N. Helberger, J. Moeller, S. Vri-
jenhoek, A. Fokkens, W. Van Atteveldt, Are we human, or are we users? The role of
natural language processing in human-centric news recommenders that nudge users to
diverse content, in: Proceedings of the 1st Workshop on NLP for Positive Impact, Asso-
ciation for Computational Linguistics, Online, 2021, pp. 47–59. doi:10.18653/v1/2021.
nlp4posimpact-1.6.
[23] N. Mattis, P. Masur, J. Möller, W. Van Atteveldt, Nudging towards news diversity: A theo-
retical framework for facilitating diverse news consumption through recommender design,
New Media & Society (2022) 146144482211044. doi:10.1177/14614448221104413.
[24] D. Friess, C. Eilders, A Systematic Review of Online Deliberation Research, Policy &
Internet 7 (2015) 319–339. doi:10.1002/poi3.95.
[25] L. M. Sanders, Against Deliberation, Political Theory 25 (1997) 347–376. doi:10.1177/
0090591797025003002.
11
Sjoerd B. Stolwijk et al. CEUR Workshop Proceedings 1–12
[26] Dahlberg, The Internet, deliberative democracy, and power: Radicalizing the public sphere,
International Journal of Media & Cultural Politics 3 (2007). doi:10.1386/macp.3.1.47/
1.
[27] S. Chambers, Rhetoric and the Public Sphere: Has Deliberative Democracy
Abandoned Mass Democracy?, Political Theory 37 (2009) 323–350. doi:10.1177/
0090591709332336.
[28] F. J. Zuiderveen Borgesius, D. Trilling, J. Möller, B. Bodó, C. H. De Vreese, N. Helberger,
Should we worry about filter bubbles?, Internet Policy Review 5 (2016). doi:10.14763/
2016.1.401.
[29] D. Owen, G. Smith, Survey Article: Deliberation, Democracy, and the Systemic Turn:
Survey Article: Deliberation & the Systemic Turn, Journal of Political Philosophy 23 (2015)
213–234. doi:10.1111/jopp.12054.
[30] J. Dryzek, Discursive democracy, Cambridge University Press, New York, 1990.
[31] H. Wessler, Habermas and the Media, Polity Press, Cambridge, UK, 2018.
[32] L. Dahlberg, The Internet and Democratic Discourse: Exploring The Prospects of Online
Deliberative Forums Extending the Public Sphere, Information, Communication & Society
4 (2001) 615–633. doi:10.1080/13691180110097030.
[33] K. Esau, D. Fleuß, S. Nienhaus, Different Arenas, Different Deliberative Quality? Using a
Systemic Framework to Evaluate Online Deliberation on Immigration Policy in Germany,
Policy & Internet 13 (2021) 86–112. doi:10.1002/poi3.232.
[34] M. Fernandez, A. Bellogin, Recommender Systems and Misinformation: The Problem or
the Solution?, OHARS Workshop. 14th ACM Conference on Recommender Systems, 22-26
Sep 2020, [Online] (2020).
[35] L. Heitz, J. A. Lischka, A. Birrer, B. Paudel, S. Tolmeijer, L. Laugwitz, A. Bernstein, Benefits
of Diverse News Recommendations for Democracy: A User Study, Digital Journalism 10
(2022) 1710–1730. doi:10.1080/21670811.2021.2021804.
[36] E. Pariser, The Filter Bubble: What the Internet is Hiding from You, Penguin, New York,
2011.
[37] C. Sunstein, Republic.Com, Princeton University Press, Princeton, NJ, 2001.
12