=Paper=
{{Paper
|id=Vol-2903/IUI21WS-SOCIALIZE-8
|storemode=property
|title=Introducing empathy into recommender systems as a tool for promoting social cohesion
|pdfUrl=https://ceur-ws.org/Vol-2903/IUI21WS-SOCIALIZE-8.pdf
|volume=Vol-2903
|authors=Alan Wecker,Tsvi Kuflik,Paul Mulholland,Belen Diaz-Agudo,Thomas Anthony Pedersen
|dblpUrl=https://dblp.org/rec/conf/iui/WeckerKMDP21
}}
==Introducing empathy into recommender systems as a tool for promoting social cohesion==
https://ceur-ws.org/Vol-2903/IUI21WS-SOCIALIZE-8.pdf
Introducing empathy into recommender systems as a tool for
promoting social cohesion
Alan Weckera, Tsvi Kuflika, Paul Mulhollandb, Belen Diaz-Agudoc and Thomas Anthony
Pedersend
a
The University of Haifa, 199 Aba Khoushy Ave, Mount Carmel, Haifa, Israel
b
The Open University, Walton Hall, Milton Keynes MK7 6AA, Great Britain
c
Instituto de Tecnología del Conocimiento, UCM, Facultad de Informática, Ciudad Universitaria, Madrid, Spain
d
Aalborg University Copenhagen, A.C. Meyers Vænge 15, Copenhagen, Denmark
Abstract
Contemporary theories of social cohesion emphasize the importance of people accepting and
appreciating differences across social groups. The SPICE project aims to promote social cohe-
sion by researching and developing tools and methods to support citizen curation for groups at
risk of exclusion. We define citizen curation as a process in which citizens can interpret cultural
objects in order to build representations of their own social group. Other groups can then engage
with those interpretations in order to appreciate their perspective. In this position paper we dis-
cuss how research into empathy can be used to motivate the design of recommender systems
that support people in looking beyond their own group and engaging constructively with alter-
native perspectives.
CCS Concepts
•Information systems applications • Collaborative and social computing systems and tools •Hu-
man-centered computing
Keywords 1
Social Cohesion, Empathy, Recommender Systems, Citizen Curation
1. Introduction p. 232). Albeit, not specifically described, what
“opportunity” means in this regard, we argue
that at a minimum it must imply an acceptance
Based on work by Pahl (1991) and Friedkin
of the other inhabitants, and as such an ac-
(2004), among others, social cohesion is argued
ceptance of the differences between oneself,
by Fonseca, Lukosch & Brazier (2018), to be
and the “others”, if not necessarily an affirma-
“[a] construct that is at the heart of what human-
tion, nor a complete understanding of these dif-
ity currently needs” (p. 231). With a specific fo-
ferences. Hence, in this view, social cohesion
cus on societies within cities, they argue that so-
can be regarded on a “higher” level, as a pinna-
cial cohesion is one of the main characteristics
cle goal of society, embracing individuality, all
of a resilient city, as “[..] fostering social cohe-
the while focusing on group unification through
sion in cities means creating societies where
the acceptance of the idiosyncrasies of the indi-
people have the opportunity to live together
vidual, the groups and the society.
with all their differences” (Fonseca et al. 2018,
Joint Proceedings of the ACM IUI 2021 Workshops, April 13-17,
2021, College Station, USA
EMAIL: ajwecker@gmail.com (A. 1); tsvikak@is.haifa.ac.il (A.
2); paul.mulholland@open.ac.uk (A. 3); belend@ucm.es (A. 4);
tape@create.aau.dk (A. 5)
ORCID: 0000-0003-4914-8949 (A. 1); 0000-0003-0096-4240 (A.
2); 0000-0001-6598-0757 (A. 3); 0000-0003-2818-027X (A. 4);
0000-0002-2193-9493 (A. 5)
Copyright © 2021 for this paper by its authors. Use permitted under Creative
Commons License Attribution 4.0 International (CC BY 4.0).
CEUR Workshop Proceedings (CEUR-WS.org)
In the SPICE project, we aim to promote so- own group but also appreciate other viewpoints
cial cohesion by researching and developing and build empathy toward those that hold them.
tools and methods to support citizen curation Empathy encompasses a number of ways in
for groups at risk of exclusion from participat- which people can respond to each other (Zaki
ing in shared culture and interacting with other 2019). These include understanding what the
groups. Groups we are working with in the other person feels (i.e. cognitive empathy),
SPICE project include older people, asylum sharing the emotion of the other person (i.e.,
seekers, children with serious illnesses, chil- emotional empathy) and wanting to improve
dren from lower socioeconomic groups, deaf the experiences of the other person (i.e., em-
people, and children from different religious pathic concern). Historically, empathy was
and secular communities. thought of as a genetic trait that operated as an
We define Citizen Curation as a process in instinct or reflex action toward other people.
which cultural objects are used as a resource by Contemporary research suggests that empathy
citizens to develop their own personal interpre- is largely environmental, and that it can change
tations (Bruni et al. 2020). Those interpreta- through life and toward different groups of peo-
tions are then shared and used within and across ple (Bazalgette 2017). In some cases, empathy
groups to reflect on similarities and differences levels can be changed relatively quickly with
in perspective. Within groups, citizens can use appropriate interventions (Zaki 2019).
their interpretations to build a representation of Currently, recommender systems are in
themselves and their shared perspective on cul- common use that aim at delivering their users
ture. Citizens from other groups can engage with relevant information. These can be partic-
with those interpretations in order to better un- ularly important in a social media context, in
derstand alternative perspectives, build empa- helping people to manage a high volume of con-
thy and thereby help to build social cohesion. tinually updated content. In our work we aim to
Citizen curation can be understood as a form investigate how empathy can be introduced into
of museum participation (Simon, 2010) in the design of recommender systems in order
which museum visitors, both physical and vir- that their users can be supported in appreciating
tual, are given opportunities to actively in en- alternative perspectives as a step toward en-
gage in culture. Social media platforms offer hancing social cohesion.
one way in which museums can promote partic-
ipation among visitors. Social media channels, 2. The Challenge: How Can Recom-
in particular Twitter, Facebook and YouTube
are commonly used by museums (Zafiropoulos mender Systems Promote Empa-
et al 2015, Badell 2015). However, analysis of thy?
museum social media accounts suggests they
are largely used for advertising rather than pub- Traditionally, recommender systems aim at
lic interaction (Badell 2015). More fundamen- assisting people in making choices without suf-
tally, although social media has the potential to ficient personal knowledge (Resnick and Var-
help people take new perspectives and interact ian 1997). Since they first appeared, in the early
with a broader range of people (Kim et al. 1990s, then called collaborative filtering sys-
2010), in practice the effects of social media are tems (Goldberg et al. 1992), they penetrated
often negative; people follow others they agree every aspect of our lives, as a means to help us-
with (homophily) (Saleem et al. 2017). This ers to cope with information overload and espe-
problem is often further exacerbated by social cially, collaborate implicitly on the task. The
media recommender systems that draw users to cultural heritage (CH) domain is just one area
people similar to themselves, sharing similar where recommender systems flourish, as
content. demonstrated by the survey of Ardissono et al.
Therefore, although social media platforms (2012). At first, recommender systems aimed
may help sub-groups to interact with each other, at recommending what seemed to be best for the
they often fail to help people to take alternative user according to the mutual taste of similar us-
perspectives. Consequently, existing social me- ers (collaborative filtering) or according to per-
dia platforms, as currently used, would not pro- sonal preferences (content-based filtering).
vide effective support for citizen curation that However, over the years, additional aspects
requires citizens to not only look within their were considered, including various contextual
aspects (Verbert et al. 2012) and more recently can have empathy toward their own group and
the idea of serendipity (Kotkov et al. 2016). a deficit of empathy toward others. Technolog-
When considering empathy as a means for en- ical developments in the 21st Century can be
hancing social cohesion, the question is how seen as accelerating the problem. Turkle (2016)
can recommender system technology can be ex- makes a link between a rapid decline in empa-
tended to consider the subtle goal of introduc- thy and ubiquitous access to digital communi-
ing empathy into its process. The first step may cations. Spinney (2017) argues that social me-
be finding a way of representing and reasoning dia can diverge the shared memories and iden-
about empathy and then including it in the rec- tities of different social groups. Can new tech-
ommendation process. When considering em- nology, and in particular recommender sys-
pathy, especially towards groups, we may find tems, increase as rather than decrease empathy?
related work in the group recommendation lit-
erature where recommendation for a group is A number of interventions can be made to
not solved as a mere aggregation of individual increase a person’s empathy toward other
preferences. For example, in the ARISE archi- groups (Bazalgette 2017, Zaki 2019). Many of
tecture (Architecture for recommendations In- these could inform the design of recommender
cluding Social Elements), Quijano et al. (2014) systems. Contact between groups can promote
proposed a recommendation method based on empathy by building understanding and an ap-
social behavior within a group including group preciation of their commonalities. Recom-
characteristics, such as size, structure, person- mender systems could suggest social contacts
ality of its members in conflict situations, and and content from other groups in order to pro-
trust between group members. Humans are so- mote cross-group contact. Perspective taking,
cial individuals and, therefore, social behavior i.e. seeing the World from someone else's per-
has a great impact on their group decision-mak- spective can promote empathy. This is particu-
ing processes. It is clear that groups have an in- larly the case if the alternative point for view is
fluence on individuals when coming to a deci- presented as a story rather than an abstract, fac-
sion. This is commonly referred to as emotional tual account (e.g. a day in the life of a home-
contagion: the effect of individuals’ affective lessness person rather than homelessness statis-
state on others in the group (Barsade 2002, Hat- tics). Evidence suggests that empathic re-
field et al. 1994, Masthoff 2004). This conta- sponses can also be strengthened if the content
gion is usually proportional to the tie strength is presented in a more intimate media such as
or trust between individuals as closer friends audio (Spence et al 2019). Recommender sys-
have a higher influence (O’Donovan and Smyth tems could prioritize content that is more per-
2005, Golbeck 2006, Victor et al. 2008). How- sonal, narrative-based and uses media such as
ever, the influence of the group also depends on audio. People tend to respond more empathi-
the individual’s degree of conformity (Masthoff cally if it is seen as a social norm. For example,
2004). It has been demonstrated that humans when reading a story by an out-group member,
adjust their opinions to conform with those of a a person is more likely to respond empathically
group when the majority of the group expresses if their peers have done the same. Recom-
a different opinion. The degree of conformity is mender systems could promote online com-
counteracted by the individual’s behavior when ments that are empathic so that this is seen as a
facing a conflict situation. Here, personality in- social norm. People also tend to respond more
fluences the acceptance of others’ proposals empathically to content if explicitly prompted
(Recio-Garcia et al. 2009) to think about the author’s point of view. Rec-
ommender systems could wrap suggested con-
People generally have higher levels of em- tent in prompts that encourage a productive re-
pathy for others from their perceived in-group. sponse. Finally, people are more likely to re-
De Waal (2011) argues that this is due to the spond empathically if they are not rushed and
tribal nature of humans (and other mammals) have the available time. Recommender systems
which was necessary for survival. People can could use contextual information (e.g. a per-
characterize their in-group in different ways, son’s current activity status) to suggest content
for example on the basis of race, gender, class, when the recipient has the time to respond em-
sexuality, religion, politics or some other char- pathically.
acteristic. Fractures between such groups create In order to promote empathy across groups,
a challenge for social cohesion, in which people the recommender system also needs a way of
identifying or constructing those groups. story is very different to Lara's interpretation of
Within the context of citizen curation, where the artwork. She adds her own comment after
visitors are supported in interpreting artworks listening.
for themselves, groups could be constructed by:
1) Social grouping i.e., explicit communities 4. Practical Challenges and Possible
based on personal attributes such as a group of
friends, or groups created based on age, sex, Solutions
race, religion; 2) Grouping based on prefer-
ences for artworks according to their attributes When considering the idea of empathy, a
(e.g. artist, subject matter, style, time period); number of practical challenges arise: How to
3) Grouping by based on the content (including reason about it? What reasoning process may
emotional content) of user interpretations pro- enable to enhance empathy towards differ-
voked by the same artwork or similar artworks. ent groups of people? How this process de-
Descriptions of artworks and emotions com- pends on the personal characteristics of the in-
bined with the use of ontologies to bring addi- dividual user? When considering the SPICE
tional meaning, provides a very rich combina- citizen curation scenario in particular, the fol-
tion of knowledge with great potential for cre- lowing practical challenges arise:
ating such communities. This type of grouping Contact: How to detect group membership
is related to the semantic similarity assessment and use this to put people in contact with other
between users. Many community detection social groups
methods have been introduced in recent years, Perspective: How to detect and recommend
with each such method being classified accord- diverse content from alternative perspectives.
ing to its algorithm type. A comprehensive re- Stories: How to detect personal, narrative-
view can be found in (Plantié and Crampes based content and prioritize for recommenda-
2013). An open research challenge is under- tion (given that it may be more empathic)
standing which type of community detection is Social norms: How to detect and prioritize
most effective for building of empathy and so- positive replies from the reader’s own social
cial cohesion. group to content from other groups?
Wrappers: How to wrap recommendations
in prompts that encourage an empathetic mind-
3. An Illustrative Scenario set? How does this relate to personality?
So, we see that empathic recommendation
The following scenario illustrates how em- requires much more than just recommending
pathy research could motivate the design of a the most appropriate content and goes beyond
recommender system. simple diversity in recommendation. It in-
Lara decides to take part in a Citizen Cura- cludes the need to reason about social groups,
tion activity on the website of her local museum. the nature of the content, social norms, and de-
The activity involves selecting an artwork from velop appropriate wrappers for presenting the
the museum's collection, adding her own inter- right content in a way that will promote empa-
pretation and sending this to a friend. She de- thy. Questions concerning ethical considera-
cides to record her interpretation as audio ra- tions also arise, including: What are considered
ther than text or video. She also chooses to legitimate methodologies to use in order to pro-
make her interpretation shareable anony- mote social cohesion via empathy and what
mously with other museum visitors. Later in the would be considered unwarranted manipula-
day when relaxing at home, Lara is notified of tions?
an interpretation of the artwork contributed by In addition, how do we measure social cohe-
someone from another social group with whom sion, in order to evaluate the success of our
she rarely interacts. The interpretation is a per- methodology? Can we measure empathy? Can
sonal story prompted by the artwork recorded we measure increases in empathy towards other
as audio. The story is accompanied by com- groups? Previous research suggests ways in
ments responding positively to the story con- which empathy can be measured. Baron-Cohen
tributed by people in Lara's social group. Lara and Wheelwright (2004) developed the Empa-
decides to listen to it. Before the audio record- thy Quotient, which is a self-report test of em-
ing starts, Lara is encouraged to imagine how pathy. Zaki (2019) reports on a number of ways
the storyteller feels about what happened. The
empathic concern can be measured from behav- modeling and user-adapted interaction,
iour such as a willingness to help someone in 22(1-2) (2012): 73-99.
need or to give to charity. Within the context of [2] Badell, J. I. “Museums and social media:
citizen curation, empathy could potentially be Catalonia as a case study.” Museum Man-
detected from the interpretations and comments agement and Curatorship, 30(3) (2015):
of visitors, for example the extent to which they 244-263.
demonstrate perspective taking. [3] Baron-Cohen, S., & Wheelwright, S. “The
Potential solutions that are considered by the empathy quotient: an investigation of
SPICE project include combining a personal adults with Asperger syndrome or high
user model with models of groups s/he may be- functioning autism, and normal sex differ-
long to. The personal user model may include ences.” Journal of autism and develop-
personal characteristics that may help a system mental disorders, 34(2) (2004): 163-175.
reason about what interests the person, together [4] Barsade SG. “The ripple effect: emotional
with personality that may guide content selec- contagion and its influence on group be-
tion and delivery. The group models may help havior.” Adm Sci Q 47(4) (2002): 644–
in selecting content that may present different 675
groups, similar or different from those the user [5] Bazalgette, P. “The empathy instinct: How
belongs to in order to cause awareness and pos- to create a more civil society”. Hachette
sibly promote empathy towards them. UK (2017).
[6] Bruni, L. E., Daga, E., Damiano, R., Diaz,
5. Conclusions L., Kuflik, T., Lieto, A., . . . Wecker, A.
“Towards Advanced Interfaces for Citizen
Curation.” Proceedings of 𝐴𝑉𝐼2𝐶𝐻 2020:
Contemporary theories of social cohesion
Workshop on Advanced Visual Interfaces
emphasize the importance of appreciating dif-
and Interactions in Cultural Heritage
ferences across social groups. Social media can
(𝐴𝑉𝐼2𝐶𝐻 2020). New York: ACM,
potentially support the sharing of alternative
(2020).
perspectives across groups. However, currently
[7] De Waal, F. “The age of empathy: Nature's
such technology often leads people toward con-
lessons for a kinder society.” Broadway
tent that fits their own viewpoint, promoting
Books, (2010).
fragmentation rather than cohesion. Research
[8] Fonseca, X., Lukosch, S., & Brazier, F.
into empathy suggests how this problem could
“Social cohesion revisited: a new defini-
be addressed by supporting people in engaging
tion and how to characterize it.” Innova-
positively with the perspectives of other groups.
tion: The European Journal of Social Sci-
We are applying this work in the cultural herit-
ence Research, 32(2) (2018): 231-253.
age domain, by developing tools and methods
[9] Friedkin, N. E. “Social cohesion.” Annual
for citizen curation, in which citizens are sup-
Review of Sociology, 30 (2004): 409-425.
ported in developing and sharing interpretations
[10] Golbeck J. “Combining provenance with
within and across social groups.
trust in social networks for semantic web
content filtering.” In: Moreau L, Foster IT
6. Acknowledgements (eds) Provenance and annotation of data,
international provenance and annotation
This project has received funding from the workshop, IPAW 2006, Chicago, IL, May
European Research Council (ERC) under the 3–5, 2006. Lecture notes in computer sci-
European Union's Horizon 2020 research and ence, vol 4145. Springer, Berlin, (2006):
innovation programme (grant agreement n° 101–108. Revised Selected Papers
870811). [11] Goldberg, D., Nichols, D., Oki, B. M., &
Terry, D. “Using collaborative filtering to
weave an information tapestry.” Commu-
7. References nications of the ACM, 35(12) (1992): 61-
70.
[1] Ardissono, L., Kuflik, T., and Petrelli, D. [12] Hatfield E, Cacioppo J, Rapson R.
“Personalization in cultural heritage: the “Emotional contagion.” Studies in emo-
road travelled and the one ahead.” User tion and social interaction. Cambridge
University Press, Cambridge, (1994).
[13] Kim, W., Jeong, O. R., & Lee, S. W. “On altering memories and changing history.”
social Web sites.” Information systems, Nature, 543, 168170, (2017)
35(2) (2010): 215-236. [26] Turkle, S. “Reclaiming conversation: The
[14] Kotkov, D., Wang, S., & Veijalainen, J. “A power of talk in a digital age.” Penguin,
survey of serendipity in recommender sys- (2016).
tems.” Knowledge-Based Systems, [27] Verbert, K., Manouselis, N., Ochoa, X.,
(2016): 111, 180-192. Wolpers, M., Drachsler, H., Bosnic, I., &
[15] Masthoff J. “Group modeling: selecting a Duval, E. “Context-aware recommender
sequence of television items to suit a group systems for learning: a survey and future
of viewers.” User Model User-Adapt In- challenges.” IEEE Transactions on Learn-
teract 14(1) (2004):37–85 ing Technologies, 5(4) (2012): 318-335.
[16] O’Donovan J, Smyth B. “Trust in recom- [28] Victor P, Cornelis C, Cock MD, Teredesai
mender systems.” In: IUI’05: proceedings A. “Key figure impact in trust-enhanced
of the 10th international conference on in- recommender systems.” AI Commun
telligent user interfaces. ACM, New York, 21(2–3) (2008):127–143
(2005): 167–174. [29] Zaki, J. “The war for kindness: Building
[17] Pahl, R. E. “The search for social cohe- empathy in a fractured world.” Broadway
sion: from Durkheim to the European Books, (2019).
Commission.” European Journal of Soci- [30] Zafiropoulos, K., Vrana, V., & Antoniadis,
ology/Archives Européennes de Sociolo- K. “Use of twitter and Facebook by top
gie, 32(2), (1991): 345-360. European museums.” Journal of Tourism,
[18] Plantié, M., and Crampes. M., Survey on Heritage & Services Marketing, 1 (1)
Social Community Detection. Springer (2015): 16-24.
Publishers. Social Media Retrieval,
(2013): 65-85.
[19] Quijano-Sanchez, L., Recio-Garcia, J. A.,
& Diaz-Agudo, B. “An architecture and
functional description to integrate social
behaviour knowledge into group recom-
mender systems.” Applied intelligence,
40(4) (2014): 732-748.
[20] Recio-García JA, Jimenez-Diaz G,
Sánchez-Ruiz AA, Díaz-Agudo B.
“Personality aware recommendations to
groups.” In: Procs of the 2009 ACM con-
ference on recommender systems. ACM,
New York, (2009): 325–328.
[21] Resnick, P., & Varian, H. R.
“Recommender systems.” Communica-
tions of the ACM, 40(3) (1997): 56-58.
[22] Saleem, H. M., Dillon, K. P., Benesch, S.,
& Ruths, D. “A web of hate: Tackling
hateful speech in online social spaces.”
arXiv preprint arXiv:1709.10159, (2017).
[23] Simon, N. “The Participatory Museum.”
Museum 2.0, Santa Cruz, CA, 2010.
[24] Spence, J., Bedwell, B., Coleman, M.,
Benford, S., Koleva, B. N., Adams, M., ...
& Løvlie, A. S. “Seeing with New Eyes:
Designing for In-the-Wild Museum Gift-
ing.” In Proceedings of the 2019 CHI Con-
ference on Human Factors in Computing
Systems, 2019: 1-13.
[25] Spinney, L. “The Shared Past That Wasn’t:
How Facebook, fake news and friends are