Singleton CTR Prediction: Towards Single-User Modeling Paradigm with Large Language Models

Click-through rate (CTR) prediction is a critical task of recommender systems. Existing CTR prediction methods typically aggregate data from multiple users, achieving significant performance but facing challenges such as privacy breaches and users' reluctance to share data. Recent advancements in large language models (LLMs) for few-shot learning have demonstrated the potential to overcome these limitations by fine-tuning with minimal data and leveraging embedded knowledge for robust performance. Motivated by this, we propose a novel paradigm-Singleton CTR prediction (SCTR)-which leverages LLMs to model CTR using data from a single user. This study aims to explore the feasibility of SCTR and identify potential research topics. To this end, we present an exploratory framework for SCTR and conduct preliminary experiments on several datasets, demonstrating the significant potential of this paradigm in personalization and privacy protection. Finally, we present some potential research tasks of SCTR for further studies.