A Network-centric Framework for Auditing Recommendation Systems

This article examines the legal questions and problems raised by the increasing use of artificial intelligence tools on social media services, in particular from the perspective of the regulations specifically governing (electronic) media. For this purpose, the main characteristics of social media services are described, and the typical forms of AI applications on social media services are briefly categorized. The analysis of the legal framework starts with the introduction of ‘protective’ and ‘facilitative’ media regulation as the two basic concepts and functions of media law in general and of the law governing information society services in particular. Against this background, the major legal challenges associated with the use of AI on social media services for both protective and facilitative media regulation are presented. With respect to protective media regulation, these challenges include the fundamental rights protection of AI-based communication on social media services, legal options to restrict such forms of communication and the responsibilities of social media providers in view of unwanted content and unwanted blocking of content. As a major objective of facilitative regulation of social media AI, the regulatory handling of potential bias effects of AI-based content filtering on social media users is discussed, including phenomena commonly referred to as ‘filter bubble’ and ‘echo chamber’ effects.

Due to their success, social network platforms are considered today as a major communication mean. In order to increase user engagement, they rely on recommender systems to personalize individual experience by filtering messages according to user interest and/or neighborhood. However some recent results exhibit that this personalization of content might increase the echo chamber effect and create filter bubbles. These filter bubbles restrain the diversity of opinions regarding the recommended content. In this paper, we first realize a thorough study of communities on a large Twitter dataset to quantify how recommender systems affect users’ behavior and create filter bubbles. Then we propose the Community Aware Model (CAM) to counter the impact of different recommender systems on information consumption. Our results show that filter bubbles concern up to 10% of users and our model based on similarities between communities enhance recommender systems.

Recommender Systems (RS) tend to recommend more popular items instead of the relevant long-tail items. Mitigating such popularity bias is crucial to ensure that less popular but relevant items are part of the recommendation list shown to the user. In this work, we study the phenomenon of popularity bias in session-based RS (SRS) obtained via deep learning (DL) models. We observe that DL models trained on the historical user-item interactions in session logs (having long-tailed item-click distributions) tend to amplify popularity bias. To understand the source of this bias amplification, we consider potential sources of bias at two distinct stages in the modeling process: i. the data-generation stage (user-item interactions captured as session logs), ii. the DL model training stage. We highlight that the popularity of an item has a causal effect on i. user-item interactions via conformity bias, as well as ii. item ranking from DL models via biased training process due to class (target item) imbalance. While most existing approaches in literature address only one of these effects, we consider a comprehensive causal inference framework that identifies and mitigates the effects at both stages. Through extensive empirical evaluation on simulated and real-world datasets, we show that our approach improves upon several strong baselines from literature for popularity bias and long-tailed classification. Ablation studies show the advantage of our comprehensive causal analysis to identify and handle bias in data generation as well as training stages.

The aim of the study is to present the online processes related to religious phenomena appearing on digital platforms, primarily the practice of content filtering (gatekeeping, echo chamber, filter bubble), and a critical review of the scientific literature on the field. At the same time, the goal is to create a theoretical introduction to the special issue and a comprehensive examination of the scientific context. For the first time, the study shows that, in terms of media content, filtering can appear from two directions. One is the selections from different events by professional journalists during content creation. The media theoretical literature refers to this aspect as the phenomenon of gatekeeping. Filtering in the other direction takes place on the part of the receivers, who choose from among the available media contents. This phenomenon has already been described by several media scholars, with the concept of selective exposure (Klapper), Daily Me (Negroponte), echo chamber (Sunstein) or filter bubble (Pariser). Focusing on the phenomenon of the filter bubble, the study presents this theory, its criticism and its relevance to religious content and religious communities. The second part of the study focuses on religious filter bubbles and presents the related investigations so far. It analyses in detail the document published by the Catholic Church on 28 May 2023, entitled Towards Full Presence, Pastoral Reflection on Engagement with Social Media. During the detailed analytical presentation of the text, the study covers how the opportunities and dangers of network communication and the use of social media appear (including the filter bubble) and what solutions the Catholic Church proposes in this regard.

This systematic review synthesizes a decade of peer-reviewed research (2015–2025) examining the interplay of filter bubbles, echo chambers, and algorithmic bias in shaping youth engagement within social media. A total of 30 studies were analyzed, using the PRISMA 2020 framework, encompassing computational audits, simulation modeling, surveys, ethnographic accounts, and mixed-methods designs across diverse platforms, including Facebook, YouTube, Twitter/X, Instagram, TikTok, and Weibo. Results reveal three consistent patterns: (i) algorithmic systems structurally amplify ideological homogeneity, reinforcing selective exposure and limiting viewpoint diversity; (ii) youth demonstrate partial awareness and adaptive strategies to navigate algorithmic feeds, though their agency is constrained by opaque recommender systems and uneven digital literacy; and (iii) echo chambers not only foster ideological polarization but also serve as spaces for identity reinforcement and cultural belonging. Despite these insights, the evidence base suffers from geographic bias toward Western contexts, limited longitudinal research, methodological fragmentation, and conceptual ambiguity in key definitions. This review highlights the need for integrative, cross-cultural, and youth-centered approaches that bridge empirical evidence with lived experiences.

In recent years, multidimensional data is becoming increasingly popular in recommender systems. For example, social tagging systems encourage users to employ user-defined keywords to help manage content in a personalized way. Recommender systems built upon social tagging systems utilize social tagging data to improve recommendation accuracy. Context-aware recommender systems incorporate context information (e.g. time, location, weather, etc.) into recommendation models to make recommendations more personalized. User profiles play a crucial role in recommender systems because user profiles provide the information about users’ information needs based on which personalized recommendations can be generated. Existing user profiling techniques for collaborative filtering (CF) recommender systems based on multidimensional data mostly analyze data through splitting the multidimensional relations into lower-dimensional relations. However, this leads to the loss of multidimensionality in user-item interactions. A different type of methods for addressing multidimensional data is factorization techniques such as tensor factorization (TF), which discover holistic latent relations between all dimensions in multidimensional data. However, TF models are not constructed to analyze any particular dimension of multidimensional data. But for many application domains, certain dimensions are of particular interest such as the user dimension in recommender systems. However, user profiles are not explicitly reflected in TF models. In this paper, we propose a unified approach to profile any particular dimension in multidimensional data based on partially-decomposed Tucker models (TM). We prove that the proposed profiling approach is a compatible higher-order extension of the matrix factorization based approaches. Moreover, we prove that the proposed TM-based profiles can be updated incrementally with new data without reconstructing the entire profiles. We integrate the proposed TF-based user/item profiling approach into neighborhood-based collaborative filtering recommenders for making top-N item recommendations. Extensive experiments have been conducted using real-world social tagging datasets to demonstrate that the CF recommenders integrated with the proposed user/item profiles outperform state-of-the-art CF recommendation approaches in terms of recommendation accuracy.

Personalization of online content creates filter bubbles and reinforces echo chambers. These are driven not only by natural human behaviours but also by design choices and efficiency-driven recommender systems. The traditional media policy goal of exposing citizens to diverse information to protect pluralism has not found its concrete application on social media. As the usage of social media as a news source increases, as well as personalization’ sophistication and group polarization, there is a need for preventing audience fragmentation. The paper suggests serendipity as a potential design principle and, eventually, policy goal. Indeed, serendipity – considered both as a capability and a process of seeking and processing unexpected and valuable information – requires diverse information as a precondition and it causes cognitive diversity. Serendipity as a design principle might encompass fundamental phases of production and consumption of information, representing a positive freedom valuable from an epistemological, psychological and political perspective. With serendipity being both limited and cultivated in the digital environment, the research reveals a theoretical trade-off between relevance and serendipity (or unknown relevance) that might be tackled with serendipity-driven recommender systems and structural and informational nudging. Such approach could turn the media policy goal of exposing users to diverse information towards an experience of diversity that comes through an architecture for serendipity.

The role social media platforms play on the emergence of polarisation is an ongoing debate in the political communication literature. Social media’s filter bubbles and online echo chambers shape people’s opinions by curating the information they have available. However, the extent to which this is the case remains unclear. Social simulation scholars have provided valuable insights into the subject through opinion dynamics models and agent-based modelling approaches. This article proposes a social simulation approach to the topic of opinion dynamics from a political communication perspective to understand how social network configurations and the media environment contribute to the emergence of national identity polarisation. We built an agent-based simulation model of national identity dynamics with a multilayer multiplex network of interacting agents in a hybrid media environment of both, traditional media and social media platforms. We use the Catalan secessionist movement to ground, contextualise and empirically inform parts of our model. We found that the initial social network setup conditions had a large impact on the emergence of polarisation amongst agents. In particular, homophily-based social networks composed of a majority of like-minded individuals produced greater polarisation compared to random networks. This was aggravated in the presence of social media filtering algorithms, selectively exposing agents to supportive information. These results emphasise the importance of both the selective exposure by social media filtering algorithms and one’s social networks (echo chambers) for polarisation to emerge. This interaction reinforces the influence of social media platforms and social networks have on the emergence of polarisation.

Meta-learning has been proven to be effective for the cold-start problem of recommender systems. Many meta-learning recommender systems that are designed for the user cold-start problem are gradient-based. They use a global parameter learned from existing users to initialize the recommender system parameter for new users that provides a personalized recommendation with limited user-item interactions. Many systems require users’ demographic information to learn the global parameter. This requirement raises privacy concerns, and demographic information is not always available. In addition, some of the gradient-based systems need dozens or even hundreds of user-item interactions from existing users to learn the global parameter. This requirement is difficult to satisfy in specific scenarios in which active users and their item interactions are rare. Moreover, gradient-based meta-learning systems rarely capture user preferences over different item attributes, as updating the corresponding weights requires many optimization steps. We proposed HyperRS, a Hypernetwork-based Recommender System, for the user cold-start problem. Our system does not rely on demographic information to provide personalized recommendations. In our system, a hypernetwork generates all weights in the underlying recommender system. The hypernetwork enables the weights to adapt quickly to capture user interest in both item attributes and the contents of the item attributes. The experimental results show that our method outperforms several state-of-the-art meta-learning recommender systems for the user cold-start problem.

Multi-attention recommender system for non-fungible tokens

Social epistemologists should be well-equipped to explain and evaluate the growing vulnerabilities associated with filter bubbles, echo chambers, and group polarization in social media. However, almost all social epistemology has been built for social contexts that involve merely a speaker-hearer dyad. Filter bubbles, echo chambers, and group polarization all presuppose much larger and more complex network structures. In this paper, we lay the groundwork for a properly social epistemology that gives the role and structure of networks their due. In particular, we formally define epistemic constructs that quantify the structural epistemic position of each node within an interconnected network. We argue for the epistemic value of a structure that we call the (m,k)-observer. We then present empirical evidence that (m,k)-observers are rare in social media discussions of controversial topics, which suggests that people suffer from serious problems of epistemic vulnerability. We conclude by arguing that social epistemologists and computer scientists should work together to develop minimal interventions that improve the structure of epistemic networks.

Recommender systems have become increasingly accurate in suggesting content to users, resulting in users primarily consuming content through recommendations. This can cause the user's interest to narrow toward the recommended content, something we refer to as preference amplification. While this can contribute to increased engagement, it can also lead to negative experiences such as lack of diversity and echo chambers. We propose a theoretical framework for studying such amplification in a matrix factorization based recommender system. We model the dynamics of the system, where users interact with the recommender systems and gradually drift'' toward the recommended content, with the recommender system adapting, based on user feedback, to the updated preferences. We study the conditions under which preference amplification manifests, and validate our results with simulations. Finally, we evaluate mitigation strategies that prevent the adverse effects of preference amplification and present experimental results using a real-world large-scale video recommender system showing that by reducing exposure to potentially objectionable content we can increase user engagement by up to 2%.

The complex interaction of humans with digitized technology has far reaching consequences, many of which are still completely opaque in the present. Technology like social networks, artificial intelligence and automation impacts life at work, at home, and in the political sphere. When work is supported by decision support systems and self-optimization, human interaction with such systems is reduced to key decision making aspects using increasingly complex interfaces. Both, algorithms and human operators become linchpins in the opaque workings of the complex socio-technical system. Similarly, when looking at human communication flows in social media, algorithms in the background control the flow of information using recommender systems. The users react to this filtered flow of information, starting a feedback-loop between users and algorithm—the filter bubble. Both scenarios share a common feature: complex human-algorithm interaction. Both scenarios lack a deep understanding of how this interaction must be properly designed. We propose the use of agent-based modeling to address the human-in-the-loop as a part of the complex socio-technical system by comparing several methods of modeling and investigating their applicability.

In the information-overloaded era of the Web, recommender systems that provide personalized content filtering are now the mainstream portal for users to access Web information. Recommender systems deploy machine learning models to learn users' preferences from collected historical data, leading to more centralized recommendation results due to the feedback loop. As a result, it will harm the ranking of content outside the narrowed scope and limit the options seen by users. In this work, we first conduct data analysis from a graph view to observe that the users' feedback is restricted to limited items, verifying the phenomenon of centralized recommendation. We further develop a general simulation framework to derive the procedure of the recommender system, including data collection, model learning, and item exposure, which forms a loop. To address the filter bubble issue under the feedback loop, we then propose a general and easy-to-use reinforcement learning-based method, which can adaptively select few but effective connections between nodes from different communities as the exposure list. We conduct extensive experiments in the simulation framework based on large-scale real-world datasets. The results demonstrate that our proposed reinforcement learning-based control method can serve as an effective solution to alleviate the filter bubble and the separated communities induced by it. We believe the proposed framework of controllable recommendation in this work can inspire not only the researchers of recommender systems, but also a broader community concerned with artificial intelligence algorithms' impact on humanity, especially for those vulnerable populations on the Web.

In an era of information explosion, recommender systems are vital tools to deliver personalized recommendations for users. The key of recommender systems is to forecast users' future behaviors based on previous user-item interactions. Due to their strong expressive power of capturing high-order connectivities in user-item interaction data, recent years have witnessed a rising interest in leveraging Graph Neural Networks (GNNs) to boost the prediction performance of recommender systems. Nonetheless, classic Matrix Factorization (MF) and Deep Neural Network (DNN) approaches still play an important role in real-world large-scale recommender systems due to their scalability advantages. Despite the existence of GNN-acceleration solutions, it remains an open question whether GNN-based recommender systems can scale as efficiently as classic MF and DNN methods. In this paper, we propose a Linear-Time Graph Neural Network (LTGNN) to scale up GNN-based recommender systems to achieve comparable scalability as classic MF approaches while maintaining GNNs' powerful expressiveness for superior prediction accuracy. Extensive experiments and ablation studies are presented to validate the effectiveness and scalability of the proposed algorithm. Our implementation based on PyTorch is available.