Understanding bias in video news & news filtering algorithms
1. capturebias.eu @CaptureBias
P1/P2:
Shared user group & data collection for
understanding bias in video news & news
filtering algorithms
Lora Aroyo, Alec Badenoch, Alessandro Bozzon, Antoaneta Dimitrova,
Markus de Jong, Claudia Hauff, Joris Hoboken, Panagiotis Mavridis,
Johan Oomen, Jesse de Vos, Claes de Vreese
2. capturebias.eu @CaptureBias
P1 FairNews Team
Nieuws voorziening in een Big Data Data tijdperk
Claes de Vreese, UvA/CW
Claudia Hauff, TU Delft
Joris van Hoboken, UvA/IvIR
Dimitrios Bountouridis, TU Delft
3. capturebias.eu @CaptureBias
P2: Capture Bias Team
Lora Aroyo, (coordinator) VU
Amsterdam, Computer Science
Alessandro Bozzon, TU
Delft CS & Delft Data
Science
Alec Badenoch, Utrecht
University, Media &
Culture Studies
Antoaneta Dimitrova,
Leiden University, Institute
of Public Administration
Markus de Jong, VU
Amsterdam Post-doc in CS
Panagiotis Mavridis,
TU Delft Post-doc in CS
& Data Science
Johan Oomen, Netherlands
Institute for Sound and
Vision
Jesse de Vos, Netherlands
Institute for Sound and
Vision
4. capturebias.eu @CaptureBias
P1/P2 Shared Goals & Vision
- P1 and P2 have shared goals and visions
- At the same time focusing on different aspects
- P1: fairness
- P2: accuracy
- Study both bias & social sorting
- explore their similarities & differences
- with respect to their manifestation in the content and social impact
- Both projects
- target news
- could share data & use groups
- We propose a two stage plan to work towards a joint demonstrator and
show added value of the synergy
5. capturebias.eu @CaptureBias
Phase 1 (year 1)
● work on a joint dataset annotated by crowd and experts
● with target annotations, e.g. entities, events, sentiment, relevance, opinions
● aim for a joint publication of this dataset as a resource paper
○ as a great reference as a benchmark dataset
○ as a basis for a research community challenge related to bias in terms of fairness and
accuracy.
● work towards discovery of alignment points between the two use cases, e.g.
○ user information needs, interaction and context of use.
○ these will be further used in the second phase.
6. capturebias.eu @CaptureBias
Phase 2 (year 2)
● work on reusing algorithms & code across the two projects
● P1 will develop recommendation algorithms for fairness in information filtering
● P2 will develop metrics for accuracy & ambiguity in bias-aware data
● validation opportunity
○ experiment with these algorithms across the two projects
○ observe empirically the interplay between fairness and accuracy
● P2 introduces also cultural and language diversity in the news sphere
○ experiments both with news video and text.
○ code sharing in Phase 2 & data sharing in Phase 1 ⇒ compare results in the different use cases
○ opportunity for another shared publication.
● P1 and P2 demonstrators with reused or shared features
7. capturebias.eu @CaptureBias
Joint Demonstrator P1/P2
● Joint dataset in Y1
○ Baselines - establishing bias parameters between viewers with diverse political views or
background through an evaluation workshops with citizens and experts
○ Github repository, python notebooks, API
● Joint user testing in Y1/Y2
○ Bias in algorithms & content
○ Accuracy & Transparency metrics and strategies
○ Diversity awareness
● Joint publications (e.g. benchmarks, resource)
○ Y1: reqs for transparency & accuracy (e.g. journalists, social sciences, media scholars)
○ Y2: results from user testing
● Joint dissemination (e.g. presentations & joint video)
○ Media Cafe (Hilversum), SPUI25, Amsterdam Data Science Meetup, Delft DS
○ Dutch Journalism Fund and CLICKNL Media & ICT events
○ Dutch ministries and NGOs events with experts from the policy analysis field
8. capturebias.eu @CaptureBias
Synergy
● Joint Dataset
○ annotated by crowds
○ validated by experts
○ Including both text & video news
● Joint Interactive Demo / Video
○ Bias-aware Data Analysis of the content
○ Bias analysis of recommender algorithms
○ Linking bias and fairness
○ Experimenting with transparency
● Partnership with related initiatives
○ at national level
○ at international level