We are pleased to share that our survey paper, “Conversational Recommender Systems Using Generative Models (Gen-CRS): A Literature Review,” has been accepted for publication in ACM Transactions on Recommender Systems (TORS).
The paper provides a comprehensive review of research on Generative Conversational Recommender Systems (Gen-CRS). As large language models and other generative technologies increasingly influence the design of conversational recommender systems, the survey examines how these developments are reshaping the field and creating new opportunities for more flexible recommendation experiences.
To structure the rapidly growing body of work, the survey organizes the literature into three analytical layers:
- Architectural and methodological foundations, covering unified, modular, and agentic system designs, approaches for adapting foundation models, and methods for recommendation and response generation.
- Knowledge and data foundations, examining how item information, user signals, and conversational data are integrated into generative recommendation workflows.
- Evaluation methodologies, reviewing what is evaluated, which quality dimensions are considered, how evaluations are conducted, who performs them, and which stakeholders are taken into account.
Beyond summarizing existing research, the survey discusses key challenges associated with generative approaches, including grounding, catalog fidelity, controllability, safety, bias, and transparency. The paper highlights the need for robust knowledge integration and comprehensive evaluation protocols that capture the complexity of conversational recommendation scenarios and support the development of reliable and trustworthy systems.
This work is the result of a collaboration between researchers from TU Wien (CDL-RecSys), the Technical University of Munich (Ashmi Banerjee), and the Polytechnic University of Bari (Fatemeh Nazary, Yashar Deldjoo).
The publication is available via the ACM Digital Library:
📄 https://dl.acm.org/doi/10.1145/3828551
