ExPerT: Personalizing LLM Responses to Users’ Domain Expertise via Query-Wise Semantic and Keystroke Behavioral Cues

Jul 16, 2026

Upload by Finn

Large language models increasingly support end users, but existing personalization techniques using static profiles or text-only signals cannot reflect how user expertise varies across individual queries. ExPerT introduces a query-wise framework that adapts LLM answers to inferred domain expertise by jointly interpreting query semantics and keystroke dynamics through in-context prompting, then generating expertise-conditioned responses controlling detail, terminology, and conceptual complexity. A study with 40 participants and 1270 queries shows 65.7% lower expertise inference error (MAE 0.398 vs 1.162) and 17.52% higher satisfaction (3.71 to 4.36, 5-point Likert).

LanguageEnglish

CategoryResearch & Report

FormatPDF

Pages35

Reading Time1 h 28 min

Introduction

Query-wise dynamic personalization

ExPerT framework overview

ExPerT: Personalizing LLM Responses to Users’ Domain Expertise via Query-Wise Semantic and Keystroke Behavioral Cues Yeji Park1 , Jiwon Tark2 , Taesik Gong1 1 UNIST, 2 Korea University [yejipark@unist.ac.kr](yejipark@unist.ac.kr) , [goneii@korea.ac.kr](goneii@korea.ac.kr) , [taesik.gong@unist.ac.kr](taesik.gong@unist.ac.kr) arXiv :2607 .01242v1 [ cs .HC] 4 May 2026 Abstract Large language models (LLMs) are increasingly used by end users, yet existing personalization methods relying on static profiles or text-only signals fail to capture query-specific expertise variation. We present ExPerT, a querywise personalization framework that adapts LLM responses to users’ query domain expertise by combining semantic and behavioral cues. ExPerT consists of two key components: (i) a semantic–behavioral expertise inference module that jointly interprets query text and keystroke dynamics via in-context LLM prompting, and (ii) an expertise-conditioned response generation that adapts the level of detail, terminology, and conceptual complexity. Our user study with 40 participants and 1270 queries demonstrated that ExPerT reduced expertise inference error by 65.7% compared to the strongest baseline (MAE = 0.398 vs. 1.162) and improved response satisfaction by 17.52%(from 3.71 to 4.36) on a 5-point Likert scale. 1 Introduction Owing to recent advances and the widespread deployment of LLMs, they are now integrated into abroad range of daily tasks, including workflow planning (Chan et al., 2025 ; Lin et al., 2025), decisionmaking processes (Eigner and Händler, 2024 ; Chiang et al., 2024), and reasoning (Wei et al., 2022 ; Wang et al., 2023a) . Consequently, users now anticipate responses that are not only accurate but also personalized to individual needs (Wanget al., 2023b), goals (Dang et al., 2025), and backgrounds (Wang et al., 2024) . To address these expectations, recent studies have focused on personalizing LLM responses by integrating user-specific text-based information, including personas (Hu and Collier, 2024 ; Sun et al., 2025), preferences (Zhao et al., 2025), and long-term user histories (Magister et al., 2024) . However, these static profile-based approaches (Hu and Collier, 2024 ; Sun et al., 2025) often fail to capture per-query dynamics, where user context can shift depending on domain expertise, task demands, or intent (Wang et al., 2024 ; Cheng et al., 2024) . To support query-wise dynamic personalization, we focus on user domain expertise—an essential factor that determines the appropriate level of detail, terminology, and conceptual complexity in LLM-generated responses (Head et al., 2021 ; Zhang et al., 2024b ; Baek et al., 2024 ; Tang et al., 2025) . Notably, a user’s expertise can vary significantly across queries, even within the same subject, making it a crucial signal for generating personalized responses (Kiseleva et al., 2015 ; Palta et al., 2025) . Yet, users’ domain expertise is difficult to infer from query text alone—short or ambiguous prompts may obscure actual expertise level (Liu et al., 2024); for instance, experts may simplify their question for efficiency, while novices may adopt technical terms without fully understanding them. These challenges motivate us to consider combining textual cues with behavioral signals to infer expertise robustly at the query level. To this end, we propose ExPerT (Expertiseaware Personalized Text generation), a query-wise response personalization framework that dynamically adapts LLM responses to a user’s query domain expertise by leveraging both semantic and behavioral cues (Figure 1) . Specifically, ExPerT introduces an expertise inference module that jointly interprets (i) query semantics, such as phrasing and domain-specific terminology, and (ii) keystroke dynamics reflecting fluency and confidence during query composition to estimate user expertise. To achieve this, we design an input feature extraction pipeline that combines query text with keystroke

FAQs

What problem does ExPerT address in LLM personalization?

ExPerT targets the limitation of existing methods that rely on static profiles or text-only cues, which fail to capture per-query changes in a user’s domain expertise. It aims to adapt response generation to that query-specific expertise level.

How does ExPerT infer a user’s domain expertise?

ExPerT combines semantic information from the query text with keystroke dynamics that reflect fluency and confidence during query composition. These cues are jointly interpreted via in-context LLM prompting to estimate expertise.

What impact did ExPerT show in the user study?

With 40 participants and 1270 queries, ExPerT reduced expertise inference error by 65.7% compared with the strongest baseline and increased response satisfaction by 17.52% on a 5-point Likert scale.

Jul 16, 2026

Upload by Finn