ZoekresultatenZoekresultaten
Volg ons op
Volg ons op
Types
0Instelling
26Publicatiejaar
6Producttype
19Bestandstype
9Taal
5Onderzoeksthema's
13Publicaties met bestand/URL
2Projectstatus
3Albeit the widespread application of recommender systems (RecSys) in our daily lives, rather limited research has been done on quantifying unfairness and biases present in such systems. Prior work largely focuses on determining whether a RecSys is discriminating or not but does not compute the amount of bias present in these systems. Biased recommendations may lead to decisions that can potentially have adverse effects on individuals, sensitive user groups, and society. Hence, it is important to quantify these biases for fair and safe commercial applications of these systems. This paper focuses on quantifying popularity bias that stems directly from the output of RecSys models, leading to over recommendation of popular items that are likely to be misaligned with user preferences. Four metrics to quantify popularity bias in RescSys over time in dynamic setting across different sensitive user groups have been proposed. These metrics have been demonstrated for four collaborative filteri ng based RecSys algorithms trained on two commonly used benchmark datasets in the literature. Results obtained show that the metrics proposed provide a comprehensive understanding of growing disparities in treatment between sensitive groups over time when used conjointly.
HOOFDSTUK
In flexible education, recommender systems that support course selection, are considered a viable means to help students in making informed course selections, especially where curricula offer greater flexibility. However, these recommender systems present both potential benefits and looming risks, such as overdependence on technology, biased recommendations, and privacy issues. User control mechanisms in recommender interfaces (or algorithmic affordances) might offer options to address those risks, but they have not been systematically studied yet. This paper presents the outcomes of a design session conducted during the INTERACT23 workshop on Algorithmic Affordances in Recommender Interfaces. This design session yielded insights in how the design of an interface, and specifically the algorithmic affordances in these interfaces, may address the ethical risks and dilemmas of using a recommender in such an impactful context by potentially vulnerable users. Through design and reflection, we discovered a host of design ideas for the interface of a flexible education interface, that can serve as conversation starters for practitioners implementing flexible education. More research is needed to explore these design directions and to gain insights on how they can help to approximate more ethically operating recommender systems.
CONFERENTIEBIJDRAGE
This report presents the design, implementation, and evaluation of a content-based recommender system for NEMO Kennislink, developed within the AI, Media and Democracy Lab (AIMD). The project investigates whether algorithmic recommendations can stimulate article discovery on a public science platform while respecting editorial and ethical principles. A content-based recommender was designed and tested in an online experiment using the Informfully research platform. Results show that generated recommendations broadened exposure patterns (i.e., increased catalogue coverage across article age, authors, and theme/series) while maintaining engagement levels comparable to editorial selections, demonstrating the potential of responsible and transparent recommender design for NEMO Kennislink.
RAPPORT
Social networks and news outlets use recommender systems to distribute information and suggest news to their users. These algorithms are an attractive solution to deal with the massive amount of content on the web [6]. However, some organisations prioritise retention and maximisation of the number of access, which can be incompatible with values like the diversity of content and transparency. In recent years critics have warned of the dangers of algorithmic curation. The term filter bubbles, coined by the internet activist Eli Pariser [1], describes the outcome of pre-selected personalisation, where users are trapped in a bubble of similar contents. Pariser warns that it is not the user but the algorithm that curates and selects interesting topics to watch or read. Still, there is disagreement about the consequences for individuals and society. Research on the existence of filter bubbles is inconclusive. Fletcher in [5], claims that the term filter bubbles is an oversimplification of a much more complex system involving cognitive processes and social and technological interactions. And most of the empirical studies indicate that algorithmic recommendations have not locked large segments of the audience into bubbles [3] [6]. We built an agent-based simulation tool to study the dynamic and complex interplay between individual choices and social and technological interaction. The model includes different recommendation algorithms and a range of cognitive filters that can simulate different social network dynamics. The cognitive filters are based on the triple-filter bubble model [2]. The tool can be used to understand under which circumstances algorithmic filtering and social network dynamics affect users' innate opinions and which interventions on recommender systems can mitigate adverse side effects like the presence of filter bubbles. The resulting tool is an open-source interactive web interface, allowing the simulation with different parameters such as users' characteristics, social networks and recommender system settings (see Fig. 1). The ABM model, implemented in Python Mesa [4], allows users to visualise, compare and analyse the consequence of combining various factors. Experiment results are similar to the ones published in the Triple Filter Bubble paper [2]. The novelty is the option to use a real collaborative-filter recommendation system and a new metric to measure the distance between users' innate and final opinions. We observed that slight modifications in the recommendation system, exposing items within the boundaries of users' latitude of acceptance, could increase content diversity.
References
1. Pariser, E.: The filter bubble: What the internet is hiding from you. Penguin, New York, NY (2011)
2. Geschke, D., Lorenz, J., Holtz, P.: The triple-filter bubble: Using agent-based modelling to test a meta-theoretical framework for the emergence of filter bubbles and echo chambers. British Journal of Social Psychology (2019), 58, 129–149
3. Möller, J., Trilling, D., Helberger, N. , and van Es, B.: Do Not Blame It on the Algorithm: An Empirical Assessment of Multiple Recommender Systems and Their Impact on Content Diversity. Information, Communication and Society 21, no. 7 (2018): 959–77
4. Mesa: Agent-based modeling in Python, https://mesa.readthedocs.io/. Last accessed 2 Sep 2022
5. Fletcher, R.: The truth behind filter bubbles: Bursting some myths. Digital News Report - Reuters Institute (2020). https://reutersinstitute.politics.ox.ac.uk/news/truth-behind-filter-bubblesbursting-some-myths. Last accessed 2 Sep 2022
6. Haim, M., Graefe, A, Brosius, H: Burst of the Filter Bubble?: Effects of Personalization on the Diversity of Google News. Digital Journalism 6, no. 3 (2018): 330–43.
ABSTRACT
This exploratory study investigates the rationale behind categorizing algorithmic controls, or algorithmic affordances, in the graphical user interfaces (GUIs) of recommender systems. Seven professionals from industry and academia took part in an open card sorting activity to analyze 45 cards with examples of algorithmic affordances in recommender systems’ GUIs. Their objective was to identify potential design patterns including features on which to base these patterns. Analyzing the group discussions revealed distinct thought processes and defining factors for design patterns that were shared by academic and industry partners. While the discussions were promising, they also demonstrated a varying degree of alignment between industry and academia when it came to labelling the identified categories. Since this workshop is part of the preparation for creating a design pattern library of algorithmic affordances, and since the library aims to be useful for both industry and research partners, further research into design patterns of algorithmic affordances, particularly in terms of labelling and description, is required in order to establish categories that resonate with all relevant parties
CONFERENTIEBIJDRAGE
This paper presents a case study about the responsible design of a recommender system of a prominent Dutch Public Service Media (PSM) organization, combining personalized content recommendations to users while aligning with the organization’s overarching mission of fostering diversity. A conceptual framework of diversity in news recommenders was translated into four possible prototypes of recommender systems, representing different ways to strike a balance between the objectives of personalization and diversity. These prototypes were presented to PSM stakeholders with different expertise, aiming to increase their insight into practical consequences of different conceptual choices, thus facilitating their communication and decision processes.
CONFERENTIEBIJDRAGE
Algorithmic affordances are defined as user interaction mechanisms that allow users tangible control over AI algorithms, such as recommender systems. Designing such algorithmic affordances, including assessing their impact, is not straightforward and practitioners state that they lack resources to design adequately for interfaces of AI systems. This could be amended by creating a comprehensive pattern library of algorithmic affordances. This library should provide easy access to patterns, supported by live examples and research on their experiential impact and limitations of use. The Algorithmic Affordances in Recommender Interfaces workshop aimed to address key challenges related to building such a pattern library, including pattern identification features, a framework for systematic impact evaluation, and understanding the interaction between algorithmic affordances and their context of use, especially in education or with users with a low algorithmic literacy. Preliminary solutions were proposed for these challenges.
CONFERENTIEBIJDRAGE
