A survey on the impacts of recommender systems on users, items, and human-AI ecosystems

TL;DR

This paper addresses the fragmentation in how recommender-system impacts are described across domains by presenting a holistic, four-ecosystem survey (social media, online retail, urban mapping, generative AI). It develops a parsimonious taxonomy of outcomes and methodologies, harmonises outcome measurements and analytical levels, and reviews data-access and ethical issues. Through 154 articles, the authors reveal broad patterns—concentration, homogenisation, diversity shifts, and volume changes—alongside system-specific dynamics such as filter bubbles and content degradation. The work provides a foundation for cross-domain comparisons, policy guidance, and design principles for responsible, human-centered recommenders, while highlighting the need for standardized data, transparency, and controlled experiments to advance understanding of the human-AI feedback loop.

Abstract

Recommendation systems and assistants (in short, recommenders) influence through online platforms most actions of our daily lives, suggesting items or providing solutions based on users' preferences or requests. This survey systematically reviews, categories, and discusses the impact of recommenders in four human-AI ecosystems -- social media, online retail, urban mapping and generative AI ecosystems. Its scope is to systematise a fast-growing field in which terminologies employed to classify methodologies and outcomes are fragmented and unsystematic. This is a crucial contribution to the literature because terminologies vary substantially across disciplines and ecosystems, hindering comparison and accumulation of knowledge in the field. We follow the customary steps of qualitative systematic review, gathering 154 articles from different disciplines to develop a parsimonious taxonomy of methodologies employed (empirical, simulation, observational, controlled), outcomes observed (concentration, content degradation, discrimination, diversity, echo chamber, filter bubble, homogenisation, polarisation, radicalisation, volume), and their level of analysis (individual, item, and ecosystem). We systematically discuss substantive and methodological commonalities across ecosystems, and highlight potential avenues for future research. The survey is addressed to scholars and practitioners interested in different human-AI ecosystems, policymakers and institutional stakeholders who want to understand better the measurable outcomes of recommenders, and tech companies who wish to obtain a systematic view of the impact of their recommenders.

Figures (3)

• Figure 1: Flowchart summarising the steps employed to construct the survey: preliminaries, where a small initial set of relevant articles was used to define a first taxonomy (box 0); paper collection, where we combined database searches, journal browsing, reference cross-checking, and expert input (box 1); filtering, where articles were screened for relevance (box 2); and categorisation, where relevant articles were assigned to one of the four ecosystems (SM, OR, UM, GAI), evaluated for relevance, and integrated into the final taxonomy (box 3).
• Figure 2: Categorisation of methodologies of the surveyed articles. At a first level, we categorise articles into empirical or simulation; at the second level, we classify them as controlled or observational.
• Figure 3: (a) Distribution of methodologies employed across ecosystems. Each cell reports the number of studies employing a methodology within an ecosystem. The bottom row reports total counts per methodology and the rightmost column reports total counts per ecosystem (170 studies in total). (b) Distribution of individual, item and ecosystem outcomes across ecosystems and methodologies. Each cell reports the number of outcomes found per ecosystem/methodology.