Problem
Research questions and friction points this paper is trying to address.
Personalized Generation
Sparse Data
Large Language Model
Cold-start Users
Reasoning-Based Personalized Generation for Users with Sparse Data
🤖 AI Summary
This study addresses the limited personalization performance of large language models under sparse interaction histories by proposing GraSPer, a novel framework that pioneers a graph reasoning-based paradigm for sparse personalized generation. By predicting future interactions and generating aligned synthetic text through reasoning, GraSPer effectively enriches user context to achieve better style and preference alignment. As the first approach integrating graph neural networks with synthetic data augmentation, it significantly enhances personalized generation in sparse scenarios across three benchmark datasets. These results validate the effectiveness of leveraging synthetic history to mitigate data insufficiency, demonstrating that structured reasoning over augmented interactions can substantially improve model adaptability even when historical user data is limited.
📝 Abstract
Large Language Model (LLM) personalization holds great promise for tailoring responses by leveraging personal context and history. However, real-world users usually possess sparse interaction histories with limited personal context, such as cold-start users in social platforms and newly registered customers in online E-commerce platforms, compromising the LLM-based personalized generation. To address this challenge, we introduce GraSPer (Graph-based Sparse Personalized Reasoning), a novel framework for enhancing personalized text generation under sparse context. GraSPer first augments user context by predicting items that the user would likely interact with in the future. With reasoning alignment, it then generates texts for these interactions to enrich the augmented context. In the end, it generates personalized outputs conditioned on both the real and synthetic histories, ensuring alignment with user style and preferences. Extensive experiments on three benchmark personalized generation datasets show that GraSPer achieves significant performance gain, substantially improving personalization in sparse user context settings.
Innovation
Methods, ideas, or system contributions that make the work stand out.
Graph-based Reasoning
Sparse Data Personalization
Context Augmentation
Reasoning Alignment
Synthetic History Generation
🔎 Similar Papers
No similar papers found.
