We are pleased to announce the program for MediaFutures first Annual Meeting, taking place in Bergen, on 29-30 September 2021. 

The goal of this year’s Annual Meeting is to give an overview and outlook of MediaFutures activities, showcase research and innovation highlights.
The Annual Meeting will bring together Norwegian and international researchers and industry participants with the intent of discussing a variety of
 key topics within media technology.

The first day of the conference is open to all interested parties. The program includes overview presentations about the MediaFutures Centre, keynote and scientific talks on media-tech topics, as well as research poster presentations by PhD candidates and graduate students. 

The second day is dedicated to industry and will be closed to MediaFutures consortium members. Participants outside the consortium are by invitation only. The program for the second day will start with an inspirational keynote talk, before we continue with an interactive session driven by industry partners. 

Advanced registration is required. There is a participation fee for most participants. 

Should you have any question regarding the conference please contact us at office@mediafutures.no.

We look forward to seeing you in September!


Day 1 – Wednesday Sept. 29

11:30 - 12:15 – Registration

Session 1: Introduction and Overview

Venue: Auditorium Atlantis, Media City Bergen

12:15 - 12:35 – Welcome Address – Prof. Christoph Trattner, Centre Director MediaFutures, University of Bergen
12:35 - 13:25 – Designing the News: datafication of news values, infrastructures and publics – Jannie Møller Hartley (Keynote)

Presented by Jannie Møller Hartley, Associate Professor & Vice-head of the interdisciplinary Centre for Big data CeBiDa, Department of Communication and Arts, Roskilde University Roskilde University.


As platforms and other algorithmic systems have become increasingly important constituents of contemporary societies, data-analytics driven practices increasingly reshape the frameworks and functions of social and economic systems. As part of this development, they also shape how journalism engages, interacts with, and cultivates its publics.

In this talk Jannie Møller Hartley explores how news production and imaginaries of the public change in the age of datafication. Drawing on interviews and ethnographic field work she surveys three four blocks of the datafication of publics namely 1) The disassembling of ‘the user’ as collective publics 2) The disassembling of the constitution of ‘the news’ 3) The algorithmic disassembling of ‘news values’ and re-assembling of algorithmic news values 4) the datafication of audiences civic engagement with the media. The talk dissects the consequences, beyond the increasingly messy relationship between news publishers and platforms and critically questions the algorithmic and data-driven shaping and cultivation of publics, and a potential reconfiguration of the very idea of a public – as Journalism moves from deciding to designing the news.


Jannie Møller Hartley is an associate professor at Roskilde University. She is currently leading the Velux funded DataPublics project, the research group Journalism and Democracy and the interdisciplinary Big Data Centre at RUC. She has been researching the digital transformation of journalism in the cross-section between audience’s use of news and journalistic news production since news media started moving online. As a former journalist she is deeply involved in the practical development of the field, from implementing open source methods in investigate journalism to the creation of ethical and value sensitive recommender systems with data scientists inside news organisations.  

Session 2: Oral Presentations

13:25 – 13:40 – Understanding Media Experiences – Brita Ytre-Arne and Hallvard Moe

Presented by Prof. Brita Ytre-Arne, University of Bergen


With the datafication of everyday life, increasingly powerful platforms and intensified competition for attention, media users face a media environment which is increasingly perceived as intrusive and exploitative of their data traces. This situation causes ambivalence and resignation as well as immersive and joyful media experiences. Understanding these experiences is crucial to democracy, as media use continues to be central for public connection and citizens’ information about and engagement in society. In addition to making sense of media usage through metrics such as clicks, time spent, shares or comments, critical attention to problematic representations of datafication should be bridged with broader and deeper understandings of media as experience.

The presentation will focus on the work being done in MediaFutures to address these challenges.


Brita Ytre-Arne is professor of Media Studies at the Department of Information Science and Media Studies at the University of Bergen, co-leader of Bergen Media Use Research Group, and Work Package Leader for WP1 Understanding Media Experiences. Her main field of research is qualitative studies of audiences and media use. She is interested in how people connect to society through cross-media user patterns, on understanding the meanings of media in everyday life, and investigating the impact of datafication, algorithms and mobile technologies on our lives. She works primarily with qualitative methods including interviews and diaries. Her current projects investigate digital disconnection dilemmas, news use during the COVID-19 pandemic, and media use and climate change. Ytre-Arne has co-authored research monographs and textbooks, edited anthologies and journal special issues, and published extensively in leading peer-reviewed journals in media and communication research. She has a substantial international network and was co-director for CEDAR – Consortium on Emerging Directions in Audience Research, a European research network that formulated an agenda for the future of audience research in the face of datafication and increasingly digital societies.

Hallvard Moe is Professor of Media Studies at the Department of Information Science and Media Studies, University of Bergen. He is also head of Bergen Media Research group with Brita Ytre-Arne, and co-leader for WP1 Understanding Media Experiences. Moe is interested in the role of media in democracy, and has researched public service broadcasting policy, and media policy more generally. He has worked on issues with media and the public sphere theoretically as well as empirically, combining a historical interest with analyses of online media from blogs to Twitter. Moe has co-authored textbooks in media studies, co-written research monographs, co-edited anthologies and journal special issues, and published around 35 articles in peer-reviewed journals as well as around 20 book chapters.

13:40 – 14:00Break

14:00 – 14:15 – User Modeling, Personalisation & Engagement – Mehdi Elahi and Christoph Trattner

Presented by Assoc. Prof. Mehdi Elahi, University of Bergen


In recent years, the ever-growing production of the media content has become a challenge for the users of media applications to find the relevant content (e.g., news articles and videos) to consume. Recommendation and personalization approaches can be potential solutions to empower media applications to support their users in discovering relevant media content and to keep them engaged. The main challenge in this context is that these approaches have little potential for the discovery of new types of media content for the users and hence might cause a number of undesired effects, e.g., the popular content becoming more popular.

In WP2 of SFI MediaFutures, we will address this challenge by developing novel models that can lead to a responsible recommendation and personalization. Such models will result in enhancing the user engagement and satisfaction by utilizing novel mechanisms, e.g., expanding recommendation diversity to cover a rich spectrum of media content and ensuring that the niche or minority content is recommended to users.


Mehdi Elahi is an Associate Professor at the University of Bergen (UiB) and the leader of WP2 at SFI MediaFutures focused on User Modeling, Personalisation & Engagement. Mehdi Elahi has obtained his Ph.D. degree in Computer Science and since then, has published more than 70 peer-reviewed journal and conference publications. His current #citation is 2100+ and his h-index is 22. His research has been mainly focused on AI, Data Science, and Cognitive Science, and their potential industrial applications such as on recommendation and personalization systems. He has co-invented and co-owned an AI-related US-patent and received prestigious research credits from big-tech IT industries (i.e., Amazon and Google). He has organized international Data Challenges together with top Companies (i.e., Spotify and XING). His research findings have been published in some of the most prestigious reference literature of the field (e.g., Recommender Systems Handbook). One of his journal articles has been the 2nd most cited paper of a top Elsevier journal.

Christoph Trattner is a Full Professor (1404) at the University of Bergen and Center Director of the Research Centre for Responsible Media Technology & Innovation – SFI MediaFutures worth around 26 million EUR and co-leader for WP2 User Modeling, Personalisation & Engagement. He is also the founder of the DARS research group at UiB and holds a 10% Research Professor (Forsker I) position in NORCE (NKLM, department of Health) one of Norway’s largest research institutions. He is also an External Lecturer at MODUL University Vienna, Austria’s leading international private university. He received a PhD (with distinction), an MSc (with distinction) and a BSc in Computer Science and Telematics from Graz University of Technology (Austria). Trattner is an ACM Senior Member and a former Austrian Research Promotion Agency (FFG) fellow, Marshall Plan and European Research Consortium for Informatics and Mathematics (ERICM) fellow and have been working at Graz University of Technology from 2009-2012, the University of Pittsburgh from 2011-2012, the Norwegian University of Science and Technology from 2014-2015, and have been visiting Yahoo! Labs Barcelona in 2014 and CWI Amsterdam in 2015 two times. I position my research in two central specializations in the Information Science research field. Since 2009, I published over 100 scientific articles in top venues about my work and have acquired over 54 million Euros in funding on European and (inter) national level – 30 million as the PI. Examples of outlets where my work has been published includes NATURE Sustainability, JASIST, EPJ Data Science, UMUAI, WWW, AAAI ICWSM or ACM SIGIR.

14:15 – 14:30 – Media Content Production & Analysis – Andreas Lothe Opdahl and Bjørnar Tessem

Presented by Prof. Andreas Lothe Opdahl and Prof. Bjørnar Tessem, University of Bergen


WP3 Media Content Production & Analysis will produce novel tools for computational journalism to produce quality generated content in terms of both trustworthiness and engagement as well as fact checking software. Central research questions are: How can we computationally produce unbiased, high-quality multi-modal content effectively? How can we analyse user-generated content accurately to generate more valuable insights? 


Andreas L. Opdahl is a Professor at the University of Bergen and work package leader of WP3 Media Content Production & Analysis. Ophdahl received his M.Eng. and Ph.D. from the Norwegian University of Science and Technology. His research interests include semantic modelling, ontologies and knowledge graphs, journalistic knowledge platforms, enterprise and IS modelling, as well as safety and security requirements. Opdahl is the author, co-author or co-editor of more than a hundred peer-reviewed research papers that have been cited several thousand times. He is a member of IFIP WG5.8 on Enterprise Interoperability and WG8.1 on Design and Evaluation of Information Systems. He serves regularly as a reviewer for premier international journals and on the program committees and as an organizer of renowned international conferences and workshops.

Bjørnar Tessem is a Professor at the University of Bergen and co-leader of WP3 Media Content Production & Analysis. Tessem holds a master in computer science (artifical intelligence, game tree search) from the University of Bergen (1985) and a PhD in computer science (artifical intelligence, Bayesian networks) from the University of Bergen (1990). He is co-leader of WP3 Media Content Production & Analysis. His research interests include artifical intelligence applied in news media analysis and content production, analogical reasoning, Bayesian networks, machine learning, and natural language understanding. Tessem has also conducted many empirical studies of software development process. He has published more than 100 scientific articles in journals, conferences, and workshops, and has led and participated in numerous projects involving artificial intelligence applied to the media and in other domains.

14:30 – 14:45 – Media Content Interaction & Accessibility – Njål Borch and Morten Fjeld

Presented by Senior Researcher Njål Borch, NORCE, and Prof. Morten Fjeld, University of Bergen


Tomorrow’s media experiences will combine sensor technology (instrumentation), AI and personal devices (interactivity) to increase engagement and collaboration. Enablers such as haptics, AR/VR, conversational AI, tangible interfaces, wearable sensors, and eyes-free interactions have made clear progress. Hence, media experiences will become more individualised, targeting the preferences and circumstances of each user (adaptation), making use of a variety of device categories offering alternative capabilities.

Research into adaptation includes responsive UIs, adaptive streaming, content adaptation and multi-device adaptation. Adaptation is also needed for collaborative and social use. Finally, media experiences must be inclusive and available for all (accessibility). Research in accessibility includes screen readers and AI-based techniques for the generation of adapted content such as audio tracks for visually impaired, video interpreters for the deaf and individualised narration.

WP4 in MediaFutures will focus on rapid prototyping and experimentation to validate approaches, ideas and user experiences. It is vital that complexity is controlled, in particular for end users, ensuring that produced knowledge is relevant for industry partners and society at large.


Njål Borch is a Senior Researcher at NORCE – Norwegian Research Centre and work package leader for WP4 Media Content Interaction and Accessibility. Borch’s main focus for research is on advanced multi-device systems, hyper-personalization, cloud based processing services and multi-angle streaming. Borch has been work package leader of several large, international research projects. Borch holds a PhD in computer science from the University of Tromsø, with a focus on distributed P2P search infrastructures.  Borch is also a founder and CEO of Motion Corporation, a company providing unique multi device synchronization services. The company started as a spin off from Norut (which is now part of NORCE). 

Morten Fjeld is a Professor at at the University of Bergen and co-leader of WP4 Media Content Interaction and Accessibility. Fjeld’s research activities are situated in the field of Human-Computer Interaction with a focus on tangible, tabletop, mobile, and collaborative computing. In 2005, he founded the t2i Interaction Lab at Chalmers, Sweden. He holds a dual MSc degree from NTNU (Norway) and ENSIMAG (France), and a PhD in HCI from ETH (Zurich). In 2002, Morten Fjeld received the ETH Medal for his PhD titled “Designing for Tangible Interaction”. In 2011, he was a visiting professor at NUS Singapore, in 2016 and 2017 at Tohoku University (Japan), and 2019 at ETH. Since 2019, he is a full professor at UiB. Morten Fjeld has extensive industrial experience in training systems, simulators, and user interface design.  

14:45 – 15:00 – Norwegian Language Technologies – Lilja Øvrelid and Koenraad De Smedt

Presented by Prof. Lilja Øvrelid, University of Oslo, and Prof. Koenraad De Smedt, University of Bergen


Language technologies are at the core of media technologies. The work package 5 in MediaFutures aims to provide datasets and models for Norwegian (Bokmål/Nynorsk) that support the automated understanding as well as the automated production of media texts in this language.

WP5 adopts theoretical approaches and methodologies primarily based on linguistic data science, including neural learning. Based on language data in the media from the user partners and data and tools at the research partners, large corpora will be annotated. The labelled examples in these corpora will be used for training and evaluating supervised models that demonstrate advanced approaches in areas such as robust deep language analysis, adaptive language generation, event identification and extraction, and analysing opinions. The partners will cooperate to explore the use of such models for innovative purposes.


Lilja Øvrelid is a professor and group leader for the Language Technology Group (LTG) within the section for Machine Learning at the Department of Informatics, University of Oslo. Øvrelid is also work package leader for WP5 Norwegian Language Technologies. Her research focuses on various types of syntactic and semantic processing of natural language text using machine learning methods, such as dependency parsing, negation/speculation analysis and sentiment analysis. This involves both the development of resources for machine learning, the development of systems which automate syntactic and semantic analysis of language, as well as the application of these systems in downstream tasks.

Koenraad de Smedt is Professor at the University of Bergen and co-leader for WP5 Norwegian Language Technologies. de Smedt studied English and Linguistics at the University of Antwerp (Belgium) where he received his degree in 1977. He became researcher, first in Antwerp and from 1983 in Nijmegen (The Netherlands) in the areas of computational linguistics, psycholinguistics and language technology. In 1985 he became lecturer at the Dept. of Experimental Psychology in Nijmegen, where he obtained his PhD in 1990 with a dissertation titled Incremental sentence generation: A computer model of grammatical encoding. In 1993 he moved to Leiden University. Since 1995 he has been professor in Computational Linguistics at the University of Bergen (Norway) and has also been affiliated part time with Unifob and Uni Research. His research interests are in psycholinguistic models of human language processing, computational linguistics, language technology, and higher education strategies.

From 1996 to 1999 he coordinated the Norwegian partnership in a European project on automatic proofreading. From 1996 to 2000 he coordinated a large scale Socrates thematic network project on Advanced Computing in the Humanities. From 2001 to 2005 he was coordinator of the Norwegian Documentation Center for Language Technology. From 2002 he coordinated three consecutive Marie Curie training projects on language resources and technologies. From 2004 to 2008 he led the Norwegian research project TREPIL on treebanking. From 2008 he participated in CLARIN and planned and coordinated the national CLARINO research infrastructure. He is currently leading the CLARINO+ project and participates in the SFI MediaFutures.

15:00 – 15:20 – Using Recommender Systems to Support Changes in Preferences and Behaviour – Alain Starke

Presented by Assoc. Prof. II Alain Starke, Wageningen University & Research / University of Bergen


Many web-based services and applications we use everyday provide personalized content. Among others, streaming platforms such as Netflix use recommender systems to present personalized content, based on one’s past viewing behaviour. The underlying algorithms often aim to optimize short-term engagement (e.g., to let users click on an advertisement), rather than serving one’s long-term preferences (e.g., to let users take up a healthier diet). This talk highlights how changing preferences and behaviour can be supported by recommender systems, by incorporating psychological principles into novel algorithms and using nudges in interface design. I will present lessons learned from the domains of energy conservation and healthy eating to suggest novel approaches in news personalization.


Alain Starke performs research at the intersection of human-computer interaction, digital nudging, and behavioral change. He obtained a PhD at Eindhoven University of Technology, the Netherlands in 2019, for his work on ‘psychologically-aware’ recommender systems in the household energy-saving domain. Afterwards, he shifted his attention more towards the food and news domains. Starke is currently a postdoctoral researcher at Wageningen University & Research, the Netherlands, where he investigates how food advice can personalized to promote healthier eating habits, by personalizing both the content and the context of the advice. This theme of adapting the ‘what’ and ‘how’ of advice is also featured in his news search and recommendation work at MediaFutures, which Starke joins in the capacity of Adjunct Associate Professor at the Department of Information Science and Media Studies, University of Bergen.

Starke has been awarded multiple grants, including two personal grants: The Research Talent Grant from the Netherlands Science Organization (€200,000) and a Niels Stensen Fellowship (€60,000). He is the main author of articles in leading journals in the energy and food domain, as well as published at CORE A-level conferences, such as CHI, IUI, and RecSys.

15:20 – 15:40 – Computational News Discovery: Designing Journalistic Tools to Support Algorithmic Lead Generation – Nicholas Diakopoulus

Presented by Assoc. Prof. Nicholas Diakopoulus, Northwesterns University / University of Bergen


Journalists are routinely challenged with monitoring vast information environments in order to identify what is newsworthy and of interest to report to a wider audience. In a process referred to as computational news discovery, alerts and leads based on data-driven algorithmic analysis can orient journalists’ attention to events, documents, or anomalous patterns in data that are more likely to be newsworthy. In this talk I will describe the development and evaluation of several computational news discovery tools that have been designed to enable journalists in domains of social, political, and investigative journalism. Findings offer insights into journalistic practices that are enabled and transformed by such news discovery tools, and suggest opportunities for improving designs to better support those practices. Implications for journalistic sourcing, sensemaking, and sustainability will be discussed. 


Nicholas Diakopoulos is an Associate Professor in Communication Studies and Computer Science (by courtesy) at Northwestern University where he is Director of the Computational Journalism Lab (CJL) and Director of Graduate Studies for the Technology and Social Behavior (TSB) PhD program. Diakopoulos’ research is in computational journalism with active research projects on (1) algorithmic accountability and transparency, (2) automation and algorithms in news production, and (3) social media in news contexts. He is the author of the award-winning book Automating the News: How Algorithms are Rewriting the Media from Harvard University Press. For some of his latest thinking and writing on automation and algorithms in journalism see his column in the Columbia Journalism Review.

15:40 – 16:00Break

Venue: Media City Bergen, Ground Floor

16:00 – 17:30 – Posters and Pitches – Media City Bergen, Ground Floor

The Poster and Pitches session takes place in Media City Bergen, Restaurant area from 16:00-17:30. The posters will be presented by MediaFutures (and affiliated) PhD candidates, Postdoctoral Researchers, Researchers and Master’s students.

The session will start with a 60 second pitch by each presenter, and the audience is invited to visit the posters thereafter.

Click the tabs for an overview.

Reaching teenagers: Public broadcasting among new media experiences

Presented by John Magnus Ragnhildson Dahl

Dodging the paywalls
Presented by Marianne Borchgrevink-Brækhus

Datafication, Media and DemocracyTransformation of news work in datafied society

Presented by Ana Milojevic

Nudging behavioral change with recommenders

Presented by Ayoub El Majjodi

How can news sites personalize people’s news experiences without making readers more polarized and fragmented?

Presented by Erik Knudsen

Exploring Recommender Systems: Towards Fair and Ethical Recommendation

Presented by Anastasiia Klimashevskaia

Large scale language models are good! But are they fair?

Presented by Samia Touileb

Fusing Media and UX

Presented by Ingar Arntzen

A brief introduction to event extraction 

Presented by Huiling You

Student Posters:

From tweet to story outline

Presented by Torstein Hatlebakk and Ulrik Wilhelm Koren

Movie recommendation based on deep visual features

Presented by David Kvasnes Olsen

The Use of Smartphones at a concert environment

Presented by Ingrid Haugsvær Årmot

Visual analysis, recommendation and personalization

Presented by Daniel C. Jakobsen

Why do people enjoy react-videos, and how come they prefer this over the original content?

Presented by Benjamin Aniket Stenerud

Building Event Graphs from GDELT Streams

Presented by Marius Alexander Pedersen


Reaching teenagers: Public broadcasting among new media experiences

John Magnus Ragnhildson Dahl, Postdoctoral Researcher, Media Futures WP1, Department of Information Science and Media Studies, UiB

Abstract: The Norwegian public broadcaster NRK has challenges in reaching a large and important group: young people. Among young people, and especially in the youth group from 10 to 19 years, NRK scores much lower both in terms of coverage, viewing time and personal importance compared to all other age groups. At the same time, we know that young people have media and technology habits that are in rapid development and are often linked to their social identity. This project therefore has two main research questions:

  1. What characterizes a selected teenage group’s media experiences, and how are they related to the needs young people experience?
  2. How do NRK’s ​​productions respond to these media experiences and needs?

The project will be carried out through three studies. In the first study, I will carry out an ethnographic fieldwork in a specific youth group. This will not be media-centered, but have media use as a key element. In the second study, I will carry out a production study at NRK in an editorial office that works with content aiming to reach young people, preferably a specigic young minority. In the third study, I will conduct a reception study using a focus group interview, where the target group in question will view and discuss the content developed NRK has made for the target group.

Dodging the paywalls

Marianne Borchgrevink-Brækhus, PhD Candidate, Media Futures WP1, Department of Information Science and Media Studies, UiB
Hallvard Moe, Professor, MediaFutures WP1 Co-Leader, Department of Information Science and Media Studies, UiB

Abstract: Although Norway remains the country with the highest number of consumers willing to pay for news, young adults are more reluctant to pay. This study analyzes news use and attitudes towards paywalls among young adults who currently don’t pay for news. Drawing on empirical data from interviews and media diaries, we ask how non-subscribers maneuver between different forms of paid and free news content. Rather than operating with a permanent distinction between subscribers and non-subscribers, the project examines how motivations for news use and payment can change over time. The study is part of the WP1 task  “Understanding future and hidden news audiences”.

Datafication, Media and DemocracyTransformation of news work in datafied society

Ana Milojevic, Postdoctoral Researcher, Marie Sklodowska-Cuire Fellow, Horizon 2020, Media Futures WP1, Media Use Research Group, Department of Information Science and Media Studies, UiB

This project investigates how audience datafication, understood as the process of converting audience interaction with media content into streams of data for computer-based processing, is transforming news work and shapes media industry. Recent studies show that audience datafication is altering news production in many respects: gatekeeping and agenda setting process; journalism practice and culture; newsroom and media organization structures; shape of media industry and media networks; journalistic role orientations and media service to society. Most of these studies revolve around the central issue which is the underlying question of this project: whether audience datafication is eroding or strengthening the democratic role of media in society. Ongoing debate about this issue has divided academic and professional community. From one point of view, audience datafication leads to production of content that attracts audience, regardless of the informative, citizen value of news. From the other, it improves connection between journalist and audience, strengthening financial base of news-work. Borrowing from sociology of news and hierarchy-of-influences theoretical approach, this project focuses on the MCB as a case study with aim to contribute to the ongoing debate by conducting research at four interrelated levels: 1) individual journalist 2) newsroom structures 3) inter-organizational and 4) reception level.

Nudging behavioral change with recommenders

Ayoub El Majjodi, PhD Candidate, Media Futures WP2, Department of Information Science and Media Studies, UiB
Christoph Trattner, Professor, MediaFutures Centre Director & WP2 Co-Leader, Department of Information Science and Media Studies, UiB
Alain Starke, Assoc. Prof. II, MediaFutures WP2, Department of Information Science and Media Studies, UiB | Postdoctoral Researcher, Wageningen University & Research

Abstract: Human preferences, subsequent decision-making and behaviors are very context dependent. Our fast and intuitive system of thinking that drives most of our decisions is prone to bias and heuristics. Furthermore, in today’s world digital tools move beyond a place for sharing and entertainment, to be an essential part of the behavioral and humanistic part of every fabric of society, such a tool confronts people to perform the right choice regarding the endless amount of information to process. Whereas recommender systems can present personalized content based on what users liked in the past, nudges can steer decision-making in an interface and other choices architectures.
In my Ph.D. project, I will examine how personalization (through recommender systems) and nudges (through social science theory) can and should be balanced to promote beneficial behaviors, such as diverse media diet.

How can news sites personalize people’s news experiences without making readers more polarized and fragmented?

Erik Knudsen, Researcher, Media Futures WP1 & WP2, Department of Information Science and Media Studies, UiB

Exploring Recommender Systems: Towards Fair and Ethical Recommendation

Anastasiia Klimashevskaia, PhD Candidate, Media Futures WP2, Department of Information Science and Media Studies, UiB

Abstract: Recommender Systems has become an integral part of various aspects of our lives, with decision making and content choosing made easier. However, in the past decade it has become more apparent that recommender systems have not only positive sides. Various drawbacks and issues have been discovered in the ways such systems effect individuals and society as whole.

In my PhD thesis I will study and attempt to address some of these undesired effects and try to help mitigate them. I will begin with one of the most well-known issues – popularity bias, which has become the downside of many algorithms in various domains.

Large scale language models are good! But are they fair?

Samia Touileb, Researcher, Media Futures WP5, Department of Information Science and Media Studies, UiB

Fusing Media and UX

Ingar Arntzen, PhD Candidate, NORCE – Norwegian Research Centre
Njål Borch, Senior Researcher, MediaFutures WP4 Leader, NORCE – Norwegian Research Centre
Anders Andersen, Professor, Head of Department of Computer Science, UiT

Abstract: Media (moving pictures) and UX (user experience) are both strong and long-standing traditions for audiovisual experiences, yet with very different technical foundations. In particular, the two traditions appear as polar opposites regarding state representation and how rendering (in time) is modelled. For example, media content is serialized into static frames on a timeline – ahead of distribution, whereas UX state may include a variety of heterogeneous, dynamic data sources and user interactivity. In media, rendering is modelled as playback along a timeline, performed by common purpose media players, while in UX, rendering is driven by application code triggering state transitions in data and programming variables.

Differences aside, there is also a strong need to combine media and UX traditions. For example, UX-based games borrow media concepts such as instant replay and employ media technologies for recording and distribution of live gameplay. Conversely, there is broad interest in integration of UX into media experiences, often to address growing demands for interactivity, adaptability, personalization, accessibility etc. This is particularly relevant for WP4 in the Media SFI.

However, due to the fundamental incompatibility between media and UX, integration is often complex and time consuming. We argue though, that this incompatibility is not given by nature, but stems from incompatibilities in fundamental concepts. Our hypothesis is that a unifying concept may be derived for media and UX, showing them as two sides of the same coin. This project aims to propose such a technical concept and evaluate relevance and implications with respect to opportunities for innovation, as well as cost saving and complexity reduction for select applications from the media and UX space.

A brief introduction to event extraction 

Presented by Huiling You, PhD Candidate, MediaFutures WP5, University of Oslo

Student Posters

From tweet to story outline

Torstein Hatlebakk and Ulrik Wilhelm Koren, Master’s Students, Media Futures WP3, Department of Information Science and Media Studies, UiB

Abstract: In our modern society of highspeed social media traffic, more events deserving of highlighting occur than journalists can keep up with. To assist in this upkeep and alleviate parts of the journalistic process, we thought it would be useful to see if it is possible to develop a pipeline for automatic generation of story outlines from social media posts. For this we will use technology from among others IMB Watson.
Such a pipeline would include technology capable of recognizing content such as text and images of social media posts such as tweets, about celebrities participating in events such as festivals, concerts, or sporting events. Then using this information, verify that the images are real, the event did actually occur and that the celebrity was present.
Additionally, the pipeline could mitigate grunt work by automatically extracting information about the celebrity that is relevant to the type of event they attended and find any other articles or stories about the topic. Lastly it could use contemporary technology to summarize, filter and present the gathered data with regards to a specific news angle.

Movie recommendation based on deep visual features

David Kvasnes Olsen, Master’s Student, Department of Information Science and Media Studies, UiB
Mehdi Elahi (Supervisor), Associate Professor, MediaFutures WP2 Leader, Department of Information Science and Media Studies, UiB
Lars Skjærven (Co-supervisor), Data Scientist, TV 2

Abstract: In recent years the movie industry has gotten an influx of new movies, which is making finding, and noticing, a “good” movie all the more difficult. Movie recommender systems are tools used to recommend movies to users based on their personal preferences. These systems observe user behavior by collecting data on the users such as likes, dislikes, reviews etc., to learn about a user’s preferences. There are two main techniques used to recommend movies to the users; collaborative and content based filtering. Collaborative filtering revolves around the similarity between user preferences, and content based filtering focuses on the similarities between the movies. The common denominator between both of these techniques is that they both require data, either about the users or the movies. Nowadays loads of movies are uploaded to platforms every day that do not have any user interaction, or suffer from a lack of item metadata such as a description or a genre tag. Thus lacking the data required to make adequate recommendations.

This thesis will focus on solving this issue by building a movie recommender system that generates recommendations based on automatically extracted “stylistic” visual features. These features can be low level features such as colorfulness or brightness, as well as higher level features such as object and emotion detection. The recommender system could help solve difficult recommendation scenarios where there is a lack of adequate data, as well as help with existing recommender systems by using new novel features that have yet to be explored.

The Use of Smartphones at a concert environment

Ingrid Haugsvær Årmot, Master’s Student, Department of Information Science and Media Studies, UiB

In present day, the smartphone holds a key position in several everyday situations. In addition to being used in the traditional way such as calling and sending text messages, it is used as a tool for practical purposes (buying bus tickets, work related, alarm clock, etc.) at the same time as it function as an entity to connect with people through social media, read the news, and document and store experiences (among several other things). In my master thesis I knew I wanted to write about something music related, and with the smartphone in mind, I quickly connected music and the smartphone to the 21th century phenomenon of documenting almost everything we might do in our lives, which includes using the smartphone to take images and film videos in a concert environment.

By conducting field observations of the concert of both Daniel Kvammen and Kjartan Lauritzen at USF Verftet in October and November, with following qualitative interviews of selected audience members, I will in my master thesis explore: How does the audience use their smartphone in a concert environment, and in which ways are the use of smartphone connected to their media experience? Is the smartphone in use when the artist is speaking, when a popular song is played or when there is a specific kind of music? Do they use the smartphone to take pictures or videos of themselves and other people around them? How do they use their stored images and videos from the concert? Do they post them to social media? All these questions would be complimented with a why in order to understand why people fill their smartphones and social media account(s) with content from a concert in Bergen.

Visual analysis, recommendation and personalization

Daniel C. Jakobsen, Master’s Student, Department of Information Science and Media Studies, UiB
Mehdi Elahi (Supervisor), Associate Professor, MediaFutures WP2 Leader, Department of Information Science and Media Studies, UiB
Igor Pipkin (Co-supervisor), Chief Data Scientist, Amedia

Personalization has become a big part of the internet, which has achieved largely using Recommender Systems. Personalization approaches exploit data from users (e.g., user clicks, ratings, etc.) and items (e.g., movie actors, directors, etc.) to create a personalized form of recommendation. Personalization can be employed in many application domains including media domain (video domain, news domain, and photo-sharing domain). In this project, the primary focus will be on Personalization of the media content utilized for advertisement in news applications.

The challenge is how to improve fairness when building a balance between the experience of users and the profit of business. In modern news applications, the user experience can be heavily disrupted by irrelevant advertisement. The solution to this problem can be building a responsible personalization through contextualization of the provided media by considering factors such as time, location, and the context of a news article. All factors can be exploited by a system to create a more appropriate experience to the user.

Understanding the factors behind fair advertisement is essential for building a responsible personalization. An example approach can be using computer vision technology to extract the visual features from the advertised media and use them to better model the content within advertised media to better filter out irrelevant content. Hence, overall aim of this project is building a responsible personalization of advertisement for both user and industry with the use of contextualization.

Why do people enjoy react-videos, and how come they prefer this over the original content?

Benjamin Aniket Stenerud, Master’s Student in Media & Communication, Department of Information Science and Media Studies, UiB

During the pandemic we lost something that may seem precious to many people. The fact that we could watch tv-series, movies and other content on YouTube together was no longer an option because we had to stay home in order to stay safe. As a product of this many streamers on the platform Twitch began to upload videos, or streams, of themselves watching another video. This has now become a trend to the point where streamers have dedicated channels to their react-videos. What’s special about this is that a lot of streamers usually sit between everything from 10 thousand to 80 thousand viewers each stream, who are watching them reacting to content, which means all of these people are actually watching the videos with the streamer. In my masters thesis i will do research on this media phenomenon and try to figure out why people watch it. Is it because they like the streamer, do they like a specific YouTube channel, or is it just more entertaining than usual. This will be done in a one-to-one interview. Therefore will my issue be about: «Why do people enjoy react-videos, and how come they prefer this over the original content?»

Building Event Graphs from GDELT Streams

Marius Alexander Pedersen, Master’s in Information Science, Department of Information Science and Media Studies, UiB

Every 15min GDELT releases a dataset with information on thousands of events collected from newspaper articles across the world. The whole dataset had a size over 2.5TB from 2020 alone, meaning this is very much “big data”. The events reported from GDELT are on a low level and often fails to capture the whole picture therefore if we manage to cluster together those that are connected, we can better capture the event, and how it evolved. The project will utilise machine learning algorithms along with word vectors and through a similarity algorithm cluster event entries in GDELT together with connected and similar events. This project builds upon work I already did this summer where I used agglomerative clustering along with a similarity scoring function between each low-level event. The similarity scoring function which used parameters as actors involved and how many articles does one event share with another will be revised and improved. One of the most difficult aspects of the project is evaluation. For as far as I can find there seems to be no gold standard on how to evaluate how correct the results are except for manually perform stick test on random samples. Then compare these samples with other news aggregators and see what is correctly clustered together along with what information on them is accurate. So, for instance see what Google News include as relevant for an event topic and compare their list with the cluster this project returns.

17:30 – 19:00Break / Individual Meetings

19:00 – Dinner, At Hotel Scandic Ørnen

Day 2 – Thursday Sept. 30 (By invitation only)

Venue: Scandic Ørnen

08:30 – 09:00Morning Coffee

Session 4: Keynote

Venue: Scandic Ørnen

09:00 – 09:50 – Natural Language Processing for Diverse News Recommendation: Challenges and Chances – Antske Fokkens (Keynote)

Presented by Dr. Antske Fokkens, Full Professor, Faculty of Humanities, Language, and at Network Institute, Vrije Universiteit Amsterdam


Quality requirements for natural language processing tools change depending on what they are used for. In this talk, I will describe our interdisciplinary approach on diversifying news recommendation. I will illustrate what the implications are when we use the NLP task of stance detection with this goal in mind. I argue that this particular use case, introduces new demands, but at the same time, allows for a simplification avoiding one of the core challenges in the tasks current setup.


Antske Fokkens is a full professor in computational linguistics at the Vrije Universiteit Amsterdam, running a chair on Computational Linguistic Methods. She obtained her PhD from Saarland University working on methodology for grammar engineering. Since she moved to the Vrije Universiteit in 2012, her main line of research has been methodology for computational linguistics in digital humanities and computational social sciences. She has worked on multiple interdisciplinary projects including BiographyNet, NewsReader and a personal VENI grant project on identifying stereotyping.

09:50 – 10:10 – Automatic News Classification – Enrico Motta

Presented by Prof. Enrico Motta, Open University / University of Bergen


Coming soon.


Professor Enrico Motta is a Professor of Knowledge Technologies at the Knowledge Media Institute (KMi) of the UK’s Open University and at the Department of Information Science and Media Studies of the University of Bergen. His work spans a variety of research areas to do with data science, semantic and language technologies, intelligent systems and robotics, and human-computer interaction. He has authored over 350 refereed publications and his h-index is 69. Over the years, Prof. Motta has received over £12M in external funding from a variety of institutional funding bodies, including EU, EPSRC, AHRC, the Arts Council, NERC, HEFCE, and Innovate UK.

His research has also attracted funding from commercial organizations, including the top two international scientific publishers, Elsevier and Springer Nature. Since 2014 he has been collaborating with Milton Keynes Council on its ‘smart city’ agenda, first as Director of the £17M, award-winning MK:Smart project, and then on the CityLabs and MK:5G initiatives.

Prof. Motta has also acted as advisor on strategic research programmes to several national and international organizations in a variety of countries, including UK, US, The Netherlands, Austria, Finland, and Estonia. He was Editor in Chief of the International Journal of Human Computer Studies from 2004 to 2018, one of the top international journals in HCI. In 2003 he founded the ground-breaking European Summer School on Ontological Engineering and the Semantic Web (SSSW), which provided the main international postgraduate forum for learning about semantic technologies until its final edition in 2016. The innovative pedagogic model pioneered in the SSSW series of summer schools has since been adopted by several other similar initiatives. 

His current activities include the use of AI techniques in the academic publishing industry; the deployment of intelligent robots in healthcare and urban settings; the development of sophisticated data management infrastructures for smart cities; and the use of AI techniques in computational journalism.

10:10 – 10:30 – The Future of Graphics in Online Live Video Production – Oskar Juhlin

Presented by Prof. Oskar Juhlin, Stockholm University / University of Bergen


Coming soon.


Oskar Juhlin is Professor at Stockholm University at the Departement of Computing and Systems Sciences. He is also Professor II at the Department of Information Science and Media Studies at Bergen University. He is founder and former Director of Mobile Life VinnExcellence Center. He has an interdisciplinary background spanning technology and social science. Oskar has conducted research and managed groups in many design fields such as fashion tech, social media and road traffic, animal computer interaction, and video interaction. His approach draws on combining ethnographic fieldwork of user practices with design and technical research, to generate knowledge and new applications, referred to as “associative design”. 

His research is mainly published at Conferences such as ACM CHI Conference on Human Factors in Computing Systems, Mobile HCI on Human-Computer Interaction with Mobile Devices and Services, and CSCW on Computer-Supported Cooperative Work and Social Computing.

Recently he has also been granted two patents and founded two companies (Figuracy/Shapecompanion AB and Liveling AB) in the areas of social media and fashion, as well as live video production.

Session 5: Table Discussions

Venue: Scandic Ørnen

10:30 – 10:40 – Introduction to World Cafe Methodology/Table Discussion

10:40 – 11:00 – Break and change table

11:00 – 11:25 – Table Discussion round 1

11:25 – 11:35 – Break and change table

11:35 – 12:00 – Table Discussion round 2

12:00 – 12:10 – Break and change table

12:10 – 12:35 – Table Discussion round 3

12:35 – 13:30Lunch with Group Photo

13:30 – 14:00 – Wrap up by Table Hosts and Centre Director

14:00 – Goodbye, safe travels!

14:00 – 16:00 – Individual WP workshops