Title:Data Watermarking for Sequential Recommender Systems

Abstract:In the era of large foundation models, data has become a crucial component for building high-performance AI systems. As the demand for high-quality and large-scale data continues to rise, data copyright protection is attracting increasing attention. In this work, we explore the problem of data watermarking for sequential recommender systems, where a watermark is embedded into the target dataset and can be detected in models trained on that dataset. We address two specific challenges: dataset watermarking, which protects the ownership of the entire dataset, and user watermarking, which safeguards the data of individual users. We systematically define these problems and present a method named DWRS to address them. Our approach involves randomly selecting unpopular items to create a watermark sequence, which is then inserted into normal users' interaction sequences. Extensive experiments on five representative sequential recommendation models and three benchmark datasets demonstrate the effectiveness of DWRS in protecting data copyright while preserving model utility.