发表机构
NLPR, Institute of Automation, Chinese Academy of Sciences; Xiaomi（中国科学院自动化研究所模式识别国家重点实验室; 小米）
机构由 AI 辅助整理，请以论文原文为准。AI 中文总结
研究个性化文本生成问题，提出无训练框架GLASS，通过全局-局部激活导向与稀疏先验实现。利用稀疏自动编码器提取全局先验、构建局部向量，推理时注入模型层。实验证明其性能优于基线，能更好分离风格与语义信息。
AI 中文摘要
个性化文本生成要求模型从历史数据中捕捉用户特定的写作风格。现有基于检索、参数高效微调或激活导向的方法，要么引入推理和存储开销，要么难以从语义内容中分离出风格信号。我们提出了GLASS，一个通过具有稀疏先验的全局-局部激活导向进行个性化生成的无训练框架。GLASS使用稀疏自动编码器从历史响应中提取全局用户风格先验，并在聚类交互场景上构建局部对比风格向量。在推理过程中，它将全局和局部向量联合注入不同模型层，实现无需检索或参数更新的上下文感知个性化。在LaMP和LongLaMP上的实验表明，GLASS在ROUGE指标和基于大语言模型的评判评估中优于基于检索、微调及导向的基线。进一步分析表明，基于稀疏自动编码器的表示对主题和长度变化更具鲁棒性，表明能更好地从语义残差中分离出风格信息。
英文摘要
Personalized text generation requires models to capture user-specific writing styles from historical data. Existing approaches based on retrieval, parameter-efficient fine-tuning, or activation steering either introduce inference and storage overhead or struggle to separate stylistic signals from semantic content. We propose GLASS, a training-free framework for personalized generation via Global-Local Activation Steering with Sparse priors. GLASS uses sparse autoencoders to extract a global user-style prior from historical responses and constructs local contrastive style vectors over clustered interaction scenarios. During inference, it jointly injects global and local vectors into different model layers, enabling context-aware personalization without retrieval or parameter updates. Experiments on LaMP and LongLaMP show that GLASS outperforms retrieval-, fine-tuning-, and steering-based baselines across ROUGE metrics and LLM-as-judge evaluations. Further analyses show that SAE-based representations are more robust to topic and length shifts, suggesting better disentanglement of stylistic information from semantic residue.
CommentsUnder review
