Autonomous decision-making systems, particularly recommender systems, have received increasing attention concerning fairness, i.e., if all stakeholders affected by such a system are treated equally as a result of the recommendations. Existing approaches primarily focus on fairness between two stakeholders – consumers and providers or consumers and items – treating providers and items as the same entity. However, we argue for the treatment of providers and items as distinct stakeholders to offer more comprehensive models of fairness in recommender systems. To this end, we propose a fairness-aware recommender system, CIPFRS, designed to optimize fairness across all three key stakeholders: consumers, providers, and items. We examine consumer fairness regarding their level of interaction with the system; high and low-activity users should be treated equally. Further, all providers should have an equal opportunity for their products to be recommended. Finally, we propose an approach to implement item fairness in each provider’s inventory. We report an extensive evaluation of the proposed solution through three datasets, demonstrating that considering all three stakeholders yields improved recommendations while minimizing bias.

Recommender systems often aim to serve multiple stakeholders, such as consumers and providers, each with distinct fairness expectations. While fairness-aware recommendation has gained increasing attention, most existing methods assume unlimited item availability. However, real-world scenarios often involve limited supply, where only a small number of item copies can be allocated. This creates new challenges in balancing fair exposure, equitable access, and relevance. In this paper, we propose a multisided fairness-aware recommendation framework designed for settings with limited item availability. Our approach explicitly models fairness both across stakeholders, ensuring consumers and providers are treated equitably compared to their peers, and within consumer–provider relationships, ensuring stakeholders treat their counterparts fairly. We formalize these as inter- and intra-stakeholder fairness and introduce evaluation metrics that measure treatment consistency under supply constraints. To address the allocation challenge, we develop a novel algorithm that assigns limited items while jointly optimizing for fairness and relevance. We evaluate our method on real-world datasets from Amazon and Goodreads, showing that it mitigates bias toward highly active users and dominant providers. Compared to conventional recommendation algorithms, our approach reduces fairness disparities by up to 80 %, underscoring the importance of fairness-aware design in real-world, resource-constrained recommendation scenarios.

Companies and organizations monitor customer satisfaction by collecting feedback through Likert scale questions and free-text responses. Freely expressed opinions, not bound to fixed questions, provide a detailed source of information that organizations can use to improve their daily operations. The organization’s quality assurance review processes require a timely follow-up on these customer opinions. However, solutions often address the analytics of textual information with topic discovery and sentiment analysis for a fixed time period. These frameworks also tend to focus on serving the purpose of a specific domain and terminology. In this study, we focus on a facilitation service to track discovered topics and their sentiments over time. This service is generic and can be applied to different domains. To evaluate the capabilities of the framework, we used two datasets with opposite types of wording. The study shows that the framework is capable of discovering similar topics over time and identifying their sentiment changes.

Nowadays, sequential recommendations are becoming more prevalent. A user expects the system to remember past interactions and not conduct each recommendation round as a stand-alone process. Additionally, group recommendation systems are more prominent since more and more people are able to form groups for activities. Subsequently, the data that a group recommendation system needs to consider becomes more complicated — historical data and feedback for each user, the items recommended and ultimately selected to and by the group, etc. This makes the selection of a group recommendation algorithm to be even more complex. In this work, we propose the SQUIRREL framework — SeQUentIal Recommendations with ReinforcEment Learning, a model that relies on reinforcement learning techniques to select the most appropriate group recommendation algorithm based on the current state of the group. At each round of recommendations, we calculate the satisfaction of each group member, how relevant each item in the group recommendation list is for each user, and based on this the model selects an action, that is, a recommendation algorithm out of a predefined set that will produce the maximum reward. We present a sample of methods that can be used; however, the model is able to be further configured with additional actions, different definitions of rewards or states. We perform experiments on three real world datasets, 20M MovieLens, GoodReads and Amazon, and show that SQUIRREL is able to outperform all the individual recommendation methods used in the action set, by correctly identifying the recommendation algorithm that maximizes the reward function utilized.

Abstract Recently, group recommendations have gained much attention. Nevertheless, most approaches consider only one round of recommendations. However, in a real-life scenario, it is expected that the history of previous recommendations is exploited to tailor the recommendations towards meeting the needs of the group members. Such history should include not only which items the system suggested, but also the reaction of the members to these items. This work introduces the problem of sequential group recommendations, by exploiting the concept of satisfaction and disagreement. Satisfaction describes how well the group received the suggested items. Disagreement describes the satisfaction bias among the group members. We utilize these concepts in three new aggregation methods, SDAA, SIAA and Average+, designed to address the specific challenges introduced by sequential group recommendations. We experimentally show the effectiveness of our methods using big real datasets for both stable and ephemeral groups.

During the last decade, the number of users who look for health-related information has impressively increased. On the other hand, health professionals have less and less time to recommend useful sources of such information online to their patients. To this direction, we target at streamlining the process of providing useful online information to patients by their caregivers and improving as such the opportunities that patients have to inform themselves online about diseases and possible treatments. Using our system, relevant and high quality information is delivered to patients based on their profile, as represented in their personal healthcare record data, facilitating an easy interaction by minimizing the necessary manual effort. Specifically, in this paper, we propose a model for group recommendations following the collaborative filtering approach. Since in collaborative filtering is crucial to identify the correct set of similar users for a user in question, in addition to the traditional ratings, we pay particular attention on how to exploit healthrelated information for computing similarities between users. Our special focus is on providing valuable suggestions to a caregiver who is responsible for a group of users. We interpret valuable suggestions as suggestions that are both highly related and fair to the users of the group. In this line, we propose an algorithm for identifying the top-z most valuable recommendations, and present its implementation in MapReduce.