=Paper=
{{Paper
|id=Vol-2758/OHARS-preface
|storemode=property
|title=Workshop on Online Misinformation- and Harm-Aware Recommender Systems: Preface
|pdfUrl=https://ceur-ws.org/Vol-2758/OHARS-preface.pdf
|volume=Vol-2758
|authors=Antonela Tommasel,Daniela Godoy,Arkaitz Zubiaga
|dblpUrl=https://dblp.org/rec/conf/recsys/X20
}}
==Workshop on Online Misinformation- and Harm-Aware Recommender Systems: Preface==
https://ceur-ws.org/Vol-2758/OHARS-preface.pdf
Workshop on Online Misinformation- and
Harm-Aware Recommender Systems: Preface
Antonela Tommasela , Daniela Godoya and Arkaitz Zubiagab
a
ISISTAN Research Institute (CONICET/UNCPBA), Tandil, Bs. As., Argentina
b
Queen Mary University of London, London, UK
Abstract
This volume contains the papers presented at the Workshop on Online Misinformation-and Harm-Aware
Recommender Systems (OHARS’2020) co-located with the 14th ACM Recommender Systems Conference
(RecSys’2020). These proceedings describe the specific workshop goals, format and contain the papers
that were presented during the online event held on September 25th, 2020.
Keywords
Recommender systems, online harms, misinformation, hate speech
1. Introduction
Misinformation and online harms are not only widespread on social media but also on other
platforms as e-commerce sites, and have shown to have serious damaging effects on individuals
and society at large [1]. Online harms include the distribution of false and misleading informa-
tion (such as hoaxes, conspiracy theories and fake news), harmful content (such as abusive or
offensive comments) and the augmentation of societal biases and inequalities online, among
others.
Recommender systems play a central role in the process of online information consumption
as they leverage massive user-generated content to assist users in finding relevant information
as well as social contacts. Thus, they are both affected by the proliferation of low-quality content
in social media, which hinders their capacity of achieving accurate predictions and, at the same
time, become unintended means for the amplification and spread of online harms.
In addition, in their attempt to deliver relevant and engaging suggestions about items, recom-
mendation algorithms might introduce biases [2], and further foster phenomena as filter bubbles
and echo chambers. Biases in data, algorithm design, evaluation, and interaction [3] limit the
exposure of users to diverse points of view and make them more vulnerable to manipulation by
disinformation [4].
OHARS 2020 was the first edition of the Workshop on Online Misinformation- and Harm-
OHARS’20: Workshop on Online Misinformation- and Harm-Aware Recommender Systems, September 25, 2020, Virtual
Event
Envelope-Open antonela.tommasel@isistan.unicen.edu.ar (A. Tommasel); daniela.godoy@isistan.unicen.edu.ar (D. Godoy);
a.zubiaga@qmul.ac.uk (A. Zubiaga)
Orcid 0000-0002-5185-4570 (D. Godoy); 0000-0003-4583-3623 (A. Zubiaga)
© 2020 Copyright for this paper by its authors. Use permitted under Creative Commons License Attribution 4.0 International (CC BY 4.0).
CEUR
Workshop
Proceedings
http://ceur-ws.org
ISSN 1613-0073
CEUR Workshop Proceedings (CEUR-WS.org)
Aware Recommender Systems1 . The aim of this workshop was to bring together researchers in
the recommender systems community interested in tackling online harms and mitigating their
impact on recommendation and, thus, to facilitate discussion about the major challenges and
opportunities that will shape future research in the area.
2. Accepted Papers
Five papers were accepted for presentation in the workshop, covering a broad range of technical
aspects related to harm-aware recommender systems, and the detection of harmful content and
its credibility.
The position paper by Fernandez and Bellogín [5] describes the key challenges behind assess-
ing and measuring the effect of existing recommendation algorithms on the recommendation of
misinforming articles and translating successful misinformation management strategies from
social science research into computational recommendation models. The authors present their
vision on how to address these problems based on four research dimensions: (1) misinformation:
problem dimensions, aims to understand what are the different dimensions of the misinforma-
tion problem, and within them, the aspects that may affect the behaviour of recommendation
algorithms; (2) analysis of recommendation algorithms, refers to the need of in-depth investi-
gations of the internal mechanisms of existing recommendation algorithms that favours the
spread of misinformation; (3) human-centred evaluation, refers to the need of modifying existing
evaluation methods and metrics to appropriately deal with misinformation; and (4) adaptation,
modification and vigilance, refers to the investigation of how recommendation algorithms could
be modified and adapted to counter their misinformation recommendation behaviour.
Ghanem et al. [6] address the detection of online users spreading hate, fake, and deceptive
online messages, usually called trolls. Text-based features, both affective and lexical ones,
coupled with topic modeling were used in this work to enhance the performance of detection
models under the assumption that they may characterize the trolls’ language changes across
topics. The study focus the experimentation on IRA (Internet Research Agency) trolls, originated
in Russia for affecting the results of the US 2016 presidential elections, and analyzes the use of
NLI (Native Language Identification) features to identify IRA trolls from their writing style.
The identification of malicious bots that degrade the performance of personalization and
recommendation algorithms on e-commerce sites with machine learning models is addressed
by Sinha et al. [7]. The authors propose two modifications to Positive-Unlabeled learning (PU
learning) framework to handle problems where the random sampling assumption is violated.
Simulation studies were conducted for validating the approaches and, then, an application of
the most general model proposed based on a large real-word dataset with traffic logs of an
e-commerce website was described.
A platform to mitigate the dangers of self-disclosure through the use of a personalized harm-
aware recommender system is proposed by Salem and Hage [8]. The recommendation algorithm
evaluates the risks of disclosing personal information and, if necessary, suggest users ways to
reduce such risks. As users are guided towards privacy preservation through nudges tailored
1
https://ohars-recsys2020.isistan.unicen.edu.ar/
according to their perception, the evaluation of the approach shows how users’ choices impact
the personalization process.
A common concern in recommendation systems is the filter bubble phenomenon, which
occurs when the system filters out information, thus narrowing the user’s perspective. This
problem is tackled by Gharahighehi and Vens [9] in the context of session-based recommenders,
which suggest the next item of interest given a sequence of previous items in the active session.
The study proposes three scenarios to make the session-based k nearest neighbor method
diversity-aware, and evaluates the scenarios across three different news datasets.
3. Program
OHARS’2020 was a half-day workshop in the context of RecSys’2020. The workshop program
included short presentations and talks with the aim of discussing on the different aspects of
harm-aware recommender systems practice and experience. The workshop started with an
opening keynote by Bárbara Poblete (University of Chile), entitled ”Mining social networks to
learn about rumors, hate speech, bias and polarization” [10], on mining social networks to learn
about rumors, hate speech, bias and polarization. The second part of the workshop started
with an invited talk by Martha Larson (Radboud University and Delft University of Technology,
Netherlands) entitled ”Moderation meets recommendation: Perspectives on the role of policies in
harm-aware recommender ecosystems” [11], on ways in which recommender systems can use
algorithms to more closely connect with moderation policies, allowing for better oversight
of system outputs and behavior. Authors of accepted submissions were invited to give a
presentations followed by some time for QA and discussion.
4. Program Committee
We would like to thank the members of the Program Committee for their valuable contribution
in providing timely and high quality reviews.
• Esma Aïmeur, Université de Montréal
• Ludovico Boratto, Eurecat
• Ivan Cantador, Universidad Autónoma de Madrid
• Giovanni Luca Ciampaglia, University of South Florida
• Leon Derczynski, University of Copenhagen
• Dagmar Gromann, University of Vienna
• Elena Kochkina, University of Warwick
• Michal Kompan, Slovak University of Technology
• Ana Maguitman, Universidad Nacional del Sur, Argentina
• Lara Quijano Sanchez, Universidad Autónoma de Madrid
• Ravi Shekhar, Queen Mary University of London
• Damiano Spina, RMIT University
• Christoph Trattner, University of Bergen, Norway
• Adam Tsakalidis, Queen Mary University of London
• Marco Viviani, University of Milano-Bicocca
• Marcos Zampieri, Rochester Institute of Technology
5. Acknowledgments
We would like to thank the RecSys 2020 workshop chairs, Jussara Almeida and Pablo Castells,
for giving us the opportunity to host this workshop and all their assistance in the workshop
organization.
The organizers are in part supported by the CONICET–Royal Society International Exchange
(IEC\R2\192019).
References
[1] C. Shao, G. Ciampaglia, O. Varol, A. Flammini, F. Menczer, The spread of fake news by
social bots, CoRR abs/1707.07592 (2017). a r X i v : 1 7 0 7 . 0 7 5 9 2 .
[2] D. Nikolov, M. Lalmas, A. Flammini, F. Menczer, Quantifying biases in online information
exposure, Journal of the Association for Information Science and Technology 70 (2019)
218–229.
[3] R. Baeza-Yates, Bias in search and recommender systems, in: Proceedings of the 14th
ACM Conference on Recommender Systems (RecSys ’20), ACM, Virtual Event, Brazil, 2020,
p. 2. doi:1 0 . 1 1 4 5 / 3 3 8 3 3 1 3 . 3 4 1 8 4 3 5 .
[4] F. Menczer, 4 reasons why social media make us vulnerable to manipulation, in: Proceed-
ings of the 14th ACM Conference on Recommender Systems (RecSys ’20), ACM, Virtual
Event, Brazil, 2020, p. 1. doi:1 0 . 1 1 4 5 / 3 3 8 3 3 1 3 . 3 4 1 8 4 3 4 .
[5] M. Fernandez, A. Bellogín, Recommender systems and misinformation: The problem or the
solution?, in: Proceedings of the Workshop on Online Misinformation- and Harm-Aware
Recommender Systems (OHARS 2020), Virtual Event, Brazil, 2020.
[6] B. Ghanem, D. Buscaldi, P. Rosso, TexTrolls: Identifying trolls on twitter with textual
and affective features, in: Proceedings of the Workshop on Online Misinformation- and
Harm-Aware Recommender Systems (OHARS 2020), Virtual Event, Brazil, 2020.
[7] S. D. R. Sinha, L. Kumari, M. Savova, Botcha: Detecting malicious non-human traffic in
the wild, in: Proceedings of the Workshop on Online Misinformation- and Harm-Aware
Recommender Systems (OHARS 2020), Virtual Event, Brazil, 2020.
[8] R. B. Salem, E. A. H. Hage, A nudge-based recommender system towards responsible online
socializing, in: Proceedings of the Workshop on Online Misinformation- and Harm-Aware
Recommender Systems (OHARS 2020), Virtual Event, Brazil, 2020.
[9] A. Gharahighehi, C. Vens, Making session-based news recommenders diversity-aware, in:
Proceedings of the Workshop on Online Misinformation- and Harm-Aware Recommender
Systems (OHARS 2020), Virtual Event, Brazil, 2020.
[10] B. Poblete, Mining social networks to learn about rumors, hate speech, bias and polarization
- Abstract, in: Proceedings of the Workshop on Online Misinformation- and Harm-Aware
Recommender Systems (OHARS 2020), Virtual Event, Brazil, 2020.
[11] M. Larson, Moderation meets recommendation: Perspectives on the role of policies in
harm-aware recommender ecosystems - Abstract, in: Proceedings of the Workshop on
Online Misinformation- and Harm-Aware Recommender Systems (OHARS 2020), Virtual
Event, Brazil, 2020.