Structuring and Tokenizing Distributed User Interest Context for Generative Recommendation

The article introduces G2Rec, a novel framework for generative recommendation that integrates holistic graph-based user co-engagement modeling with semantic tokenization. G2Rec addresses scalability issues and the limitations of existing methods by capturing comprehensive user interest contexts without relying on ground-truth user interests, improving the accuracy of user behavior modeling. Its deployment and extensive experiments on public datasets indicate superior performance compared to traditional approaches, making it relevant for practitioners seeking to enhance recommendation systems in industrial applications.