Generative Archetype-Grounded Item Representations for Sequential Recommendation

The paper introduces GenAIR, a framework for enhancing sequential recommendation systems by using Generative Archetype-grounded Item Representations. It employs a large language model to generate item archetypes based on metadata, and incorporates a behavioral calibration objective to align the embedding space with real user interactions, leading to improved performance across various models on three datasets. This approach addresses the limitations of static item representations and enhances the semantic understanding of user behavior, making it a valuable tool for practitioners in the recommendation domain.