PerDucer: Keyphrase-Driven Personalization Inducer for Summarization from User Histories

Document summarization becomes more challenging when summaries must reflect a user's subjective interests in addition to document salience. SOTA Large Language Models (LLMs) show strong in-context summarization capabilities. Prior works report that simply prepending long and dynamically evolving user histories leads to unstable, inconsistent personalization. To address this, we introduce PerDucer, a personalization inducer for frozen language models. Given a user interaction sequence (trajectory) and a query document, PerDucer first predicts the next likely preference signal. It then maps the latent signal to a small set of personalized keyphrases for the query document. These keyphrases serve as the control cues that steer the frozen summarizers (both LLMs and SLMs) towards user-aligned summaries. Across PENS and a reconstructed temporal version of OpenAI-Reddit, PerDucer-boosted LLMs outperform the strongest history-prompting baselines, yielding an average +0.18 improvement across personalization metrics (PerSEval). Two PerDucer-augmented SLMs approach the topperforming evaluated LLM, with SmolLM2-1.7B reaching ≈97% of the best-performing DeepSeek-R1-14B score. These results suggest that short keyphrase cues can induce personalization in frozen summarizers without modifying their parameters. The codes are available at https://github.com/KDM-LAB/ PerDucer-ACL_2026_Findings.git .