Contextual information has been widely recognized as an important modeling dimension in various social science and technological disciplines and is becoming more and more important for enhancing recommendation results and retrieval performance.
There have been several CARS workshops organized in the past, including a successful workshop that was held last year with more than 100 participants, where the addition of contextual information to traditional recommender systems has been discussed. While a substantial amount of research has already been performed, many existing approaches to context-aware recommender systems (CARS) focus on the so-called 'representational view' that incorporates pre-defined and static contextual factors (such as time and location) in the recommendation process. Furthermore, to capture implicit context patterns, deep learning approaches can extract latent contextual representations using embedding in low-dimensional spaces. These neural embeddings enable recommender systems to identify latent contextual factors that traditional methods might overlook, adapting to evolving contextual dimensions through transformer architectures and self-supervised learning techniques.
Discovering contextual information from multiple types of data (semantic web, graphs) and media (text, images, video, speech) can enhance recommendation quality. Recent research has also demonstrated that modeling context as a latent space may address the sparsity and high dimensionality challenges inherent in CARS models. For example, latent context-aware recommender systems utilize unsupervised learning techniques to infer implicit contextual information from mobile devices. Other recent studies have shown that sequential contextual information improves recommendation accuracy, as sequences enable modeling both long- and short-term user preferences. Furthermore, recent work on explainable CARS and fairness-aware CARS has introduced models that increase transparency and user trust in recommendations.
The increasing adoption of recommender systems in mobile applications, IoT environments, and multi-stakeholder settings further emphasizes the need for novel context-aware modeling strategies. For example, privacy and security concerns in context-aware recommendations necessitate robust modeling approaches, such as federated learning, to preserve user data while maintaining recommendation quality. Additionally, ethical considerations in CARS, including fairness and bias mitigation, remain important challenges.
The primary goal of the CARS workshop is to rethink the role of contextual information in recommender systems and broadly discuss key features of the next generation of CARS. The workshop aims to explore application domains that require novel contextual information types and cope with their dynamic properties. In this respect, the main challenge for the next generation of CARS is to introduce more explainable, flexible, and comprehensive approaches to modeling and utilizing contextual information. Additionally, as privacy and security risks become increasingly prominent in CARS applications, the workshop will also address privacy-preserving techniques, secure computation methods, and federated approaches to safeguard user data while ensuring effective recommendations. The workshop seeks to discuss novel perspectives on how recommender systems can adapt to diverse contextual situations, bringing together researchers with wide-ranging expertise to identify key research questions, exchange ideas across disciplines, and foster innovation in the next generation of context-aware recommender systems.
Topics of interest
We invite contributions to the workshop about topics related to CARS (but are not limited to):
· Context in generative recommender systems
· Explainable context-aware recommender systems
· Understanding user intent and dynamic preferences for improved personalization
· Foundation Models for Contextual Information
· Privacy and security modeling in CARS
· Federated Learning for CARS
· Ethics and Fairness in CARS
· Context in Augmented and Virtual Reality
· Context Processing in IoT Environments
· Sequence-aware and time-aware recommender systems
· Latent context models for recommender systems
· GUI strategies for increasing context awareness
· Data sets for context-dependent recommendations
· Human context recognition for health applications
Submission Types and Guidelines
CARS submissions should be prepared in PDF format according to the new single-column format (Microsoft Word or Latex formats). If you are using Overleaf, you can use the following code (\documentclass[manuscript]{acmart}). The peer-review process is single-blind and handled electronically through EasyChair. Accepted papers will be included in the workshop proceedings and at least one author of each accepted contribution must attend the workshop. Accepted papers are given an oral or a poster presentation slot at the workshop.
The ideal length of a paper for the CARS workshop is between 4-8 pages (excluding references). Submitted work should be original. However, technical reports or ArXiv disclosure prior to or simultaneous with the workshop submission is allowed, provided they are not peer-reviewed. The organizers also encourage authors to make their code and datasets publicly available.
Paper submission deadline: July 10, 2025
Notification: August 6th , 2025
Camera-ready deadline: August 20th, 2025
Workshop (at RecSys 2025): TBD