GNN-based text representations for Style-focused Personalized LLM’s

Abstract: Large Language Model (LLM) personalization is the task of making the text that an LLM generates similar to the text written by a certain user. In recent years, this task has become increasingly important with a wide range of applications. However, the topic of evaluating the quality of the personalization is largely unexplored. In this study, we introduce a novel approach to learning style-focused text representations using graph neural networks (GNNs) that can be used to evaluate LLM personalization. In addition, we show how these representations can be used to select demonstrations that can be used through in-context learning (ICL) to personalize the LLM's generated text.