The core insight driving the special issue is deceptively simple: recommender systems do not merely predict preferences, they actively shape consumption, information exposure, mobility and collective behavior. By deciding which products, services, content, destinations or actions are presented to users, these algorithms can influence resource consumption and environmental impact, while also creating opportunities to promote more sustainable choices. The editorial distinguishes two related strands of work that have emerged in the literature. Green recommender systems traditionally emphasize environmental aspects such as reducing energy consumption, carbon emissions and other ecological impacts, while the broader movement of recommender systems for good focuses on positive societal effects more generally. The notion of recommender systems for sustainability, the editors argue, encompasses both perspectives and connects them to a multidimensional understanding of sustainability spanning environmental, economic and social dimensions.

A crucial and often overlooked dimension of the debate concerns the environmental cost of the algorithms themselves. The editorial points out that the development and operation of modern recommender systems require substantial computational resources, particularly with the increasing use of large-scale machine learning and foundation models. Recent work on green recommender systems has begun quantifying and minimizing the carbon footprint of AI-powered personalization. Sustainability, in other words, must be considered twice: as a property of the outcomes produced by recommender systems, and as a property of the systems and infrastructures used to generate recommendations. This dual role creates genuine tension, because optimizing purely for engagement or commercial performance can encourage excessive consumption, reinforce inequalities and increase computational and environmental costs.

The accepted papers illustrate how varied and technically sophisticated this field has become. Yen and Phuong contribute RETSRec, a sequential recommendation architecture that combines behavioral sequences with textual reviews to capture both interaction patterns and the semantic information underlying user preferences. By integrating a bidirectional transformer with a BERT-based review encoder, the model achieves significant improvements over state-of-the-art methods on real-world Amazon datasets, demonstrating that richer preference understanding can serve more intentional, less wasteful consumption. Farrokhizhan and Öztayşi take a different route, introducing a two-phase framework that combines hesitant fuzzy user clustering with adaptive association rule mining and incorporates SDG-aligned eco-scores into the recommendation utility function. On a fashion e-commerce dataset, their sustainability-aware variant substantially outperforms the baseline while maintaining comparable computational requirements, a result the editors highlight as evidence that greener recommendations need not come at the price of accuracy or efficiency.

Evaluation methodology receives particular attention. Felfernig and colleagues propose sustainability-aware evaluation metrics that extend conventional measures such as accuracy, precision, recall and user satisfaction by incorporating the environmental, social and economic impacts of recommender systems in alignment with the Sustainable Development Goals. Through a case study on a real-world product dataset, they demonstrate how such metrics can reveal broader and longer-term impacts of recommendation systems that remain hidden when using conventional performance measures alone. This methodological shift matters because what a community measures ultimately determines what it builds; if only click-through rates are optimized, sustainability consequences stay invisible. In a related contribution, Yang and Geng present DRPRCDA, an end-to-end framework combining graph-based representation learning with bias-aware reinforcement learning to dynamically balance recommendation accuracy and diversity while mitigating popularity bias, showing that the dominance of popular items can be reduced without sacrificing performance.

Fairness and feedback dynamics form a second technical pillar of the issue. Nguyen benchmarks eight recommender systems, from classical matrix factorization to neural and graph-based models, using both accuracy and fairness metrics on two MovieLens datasets, finding that high-performing models can also deliver competitive fairness. Zoralioglu and Yalcin go deeper into the temporal dimension, investigating how iterative feedback loops influence fairness, calibration, accuracy, diversity and popularity bias for users with popular, diverse or niche preferences. Their findings are sobering: feedback loops progressively reinforce structural inequalities, with niche-focused users experiencing the greatest deterioration in calibration, diversity and long-tail exposure, while users who begin with diverse tastes gradually converge toward popularity-driven recommendations. From a sustainability perspective, such dynamics matter because they can concentrate consumption patterns and squeeze out niche, potentially more sustainable alternatives.

Several contributions extend recommendation technology into domains with direct sustainability stakes. Yildiz and Bitirim present RERS, a real-time emotion-aware recommender system for sustainable and eco-friendly driving that combines physiological sensing, emotion recognition, mobile technologies and a recommendation agent to encourage greener driving while supporting driver well-being, achieving an overall emotion-recognition accuracy of 75.01 percent. Ye and colleagues tackle travel mode choice with METMP-LSTMA, a model combining LSTM networks with attention mechanisms for capturing temporal travel preferences and Model-Agnostic Meta-Learning for addressing cold-start users with limited historical data; experiments on two real-world datasets show an average accuracy improvement of 5.7 percent over baselines and a striking 10.9 percent improvement for cold-start users. Ghahramani and co-authors propose a hybrid tourism recommender that integrates geo-tagged photographs and metadata with textual reviews and demographic information using content-based, collaborative and demographic filtering, with the Bat Algorithm optimizing feature weighting and geographical clustering; experiments on Yelp data demonstrate improvements in precision, recall, F1, diversity and novelty.

The issue also reaches into agriculture, healthcare and organizational efficiency. Chetan and colleagues develop an ensemble-based recommender system for sustainable agriculture supporting both crop selection and fertilizer recommendation, using local interpretable model-agnostic explanations to make predictions transparent while identifying important agricultural factors such as moisture, phosphorus and soil type. Two contributions target traditional Chinese medicine: Zhang and colleagues introduce TCM_DiffPR, a knowledge-graph diffusion model incorporating personalized patient attributes, prompt fine-tuning, contrastive learning and collaborative knowledge-graph signals for personalized herbal prescription recommendation, while Zhang and Tan separately propose interpretable methods combining graph neural networks, attention mechanisms and a manually constructed knowledge graph, explicitly emphasizing technical and social sustainability in transparent, personalized diagnosis support. On the efficiency front, Ram, Lai and Xiao present a workflow recommendation framework combining singular value decomposition with machine-learning-based clustering to learn sequential application usage patterns from a no-code automation platform, consistently outperforming Markov-chain and random strategies while reducing cognitive load and supporting organizational well-being.

Underpinning several of these systems is a new generation of large language model architectures. Xing and colleagues introduce AlignGenRec, which aligns collaborative-filtering item embeddings with textual representations from item descriptions before transferring the combined knowledge to a large language model for generative recommendation, outperforming both conventional collaborative-filtering and LLM-based baselines particularly in cold-start settings. The editors note that such powerful models bring their own sustainability challenge, given the substantial computational resources required for training and inference. Yet the breadth of the fourteen accepted papers, spanning transformers, diffusion models, reinforcement learning, meta-learning, fuzzy clustering and metaheuristic optimization, suggests a community capable of addressing both sides of the equation: making recommendations themselves greener and making what they recommend more sustainable.

The editors conclude that the strong interest in this special issue, reflected in a selective acceptance rate of approximately 21 percent, illustrates the growing importance of these questions for the recommender systems community. Their central message is that recommendations should not only be accurate and useful but may also need to account for their consequences for individuals, society and the planet. Recommender systems, they argue, can potentially contribute to all of the United Nations Sustainable Development Goals, yet the same technologies can create tensions with sustainability objectives when engagement optimization encourages overconsumption or increases environmental costs. By bringing together research on green recommender systems, recommender systems for good and related approaches under a single sustainability umbrella, this special issue establishes a common agenda for a field that increasingly recognizes a sobering truth: the algorithms that decide what billions of people see, buy and do are no longer neutral infrastructure, but a powerful lever that must be pointed deliberately toward a sustainable future.

Subject of Research: Recommender systems for sustainability

Article Title: Special Issue: Recommender systems for sustainability

Article References: Felfernig, A., Helic, D., Tran, T. N. T., & Calero Valdez, A. (2026). Special Issue: Recommender systems for sustainability. Journal of Intelligent Information Systems. https://doi.org/10.1007/s10844-026-01091-2

Image Credits: AI Generated

DOI: 10.1007/s10844-026-01091-2

Keywords: recommender systems, sustainability, green recommender systems, recommender systems for good, Sustainable Development Goals, fairness, popularity bias, carbon footprint, personalization, evaluation metrics, machine learning, sustainable consumption