U-Lens: 支持长篇 LLM 回答中的用户不确定性管理

Jul 16, 2026

Upload by Riley

Large language models increasingly generate long-form answers, yet users struggle to determine what in a response should be scrutinized, why claims may be unreliable, and what to do next. Instead of only making uncertainty visible, the work studies user uncertainty management across interpretation, evaluation, and decision stages, then derives design guidelines covering uncertainty target representation, evaluative explanation, response guidance, and interactive presentation. U-Lens implements these ideas and improves verification efficiency, effort allocation, perceived workload, and perceived support versus a confidence-cue baseline.

LanguageEnglish

CategoryResearch & Report

FormatPDF

Pages24

Reading Time1 h 0 min

Introduction

User uncertainty management in long-form LLM responses

Research questions and study stages

Design guidelines and system overview

U-Lens: an uncertainty-management support system

Uncertainty inspection targets and prioritization

Evaluative context and response options

Evaluation and results

Controlled within-subject study setup

Comparative outcomes versus confidence cues

U-Lens: Supporting User Uncertainty Management in Long-Form LLM Responses Yu Mei Tsinghua University Beijing, China [meiy24@mails.tsinghua.edu.cn](meiy24@mails.tsinghua.edu.cn) Chang Liu Tsinghua University Beijing, China [c-liu21@tsinghua.org.cn](c-liu21@tsinghua.org.cn) Qingyue Zhuang Tsinghua University Beijing, China [zhuangqy23@mails.tsinghua.edu.cn](zhuangqy23@mails.tsinghua.edu.cn) Zhi Zheng Tsinghua University Beijing, China [zhengz22@mails.tsinghua.edu.cn](zhengz22@mails.tsinghua.edu.cn) Jie Cai Tsinghua University Beijing, China [jie-cai@mail.tsinghua.edu.cn](jie-cai@mail.tsinghua.edu.cn) Zhoutong Ye Tsinghua University Beijing, China [yezt24@mails.tsinghua.edu.cn](yezt24@mails.tsinghua.edu.cn) Chun Yu Tsinghua University Beijing, China [chunyu@tsinghua.edu.cn](chunyu@tsinghua.edu.cn) Yuanchun Shi Tsinghua University Beijing, China [shiyc@tsinghua.edu.cn](shiyc@tsinghua.edu.cn) arXiv :2607 . 10604v1 [ cs .HC] 12 Jul 2026 Abstract Large language models (LLMs) are increasingly used to generate long-form answers for knowledge-intensive tasks, but users often struggle to decide which parts of a response deserve scrutiny, why they may be unreliable, and what to do next. Prior work on uncertainty communication has largely focused on making uncertainty visible through cues such as confidence scores, leaving less support for the broader process of managing uncertainty distributed across a long response. Through a formative study, we examine how users manage such uncertainty across three stages: interpretation, evaluation, and decision. Based on these insights, we derive design guidelines that address both stage-specific and cross-stage needs: uncertainty target representation, evaluative explanation, response guidance, and interactive presentation. We instantiate these guidelines in U-Lens, an uncertainty-management support system that organizes uncertain information in long-form responses into contextual inspection targets, prioritizes them for attention, and connects each target with evaluative context and response options. We evaluated U-Lens in a controlled within-subjects study with 18 participants, comparing it against a confidence-cue baseline. Our results show that U-Lens improved verification efficiency and effort allocation, lowered perceived workload, and strengthened perceived support across interpretation, evaluation, and decision stages. This work reframes uncertainty support for generative AI from presenting isolated, text-centered cues toward supporting the user-centered process of interpreting, evaluating, and acting on uncertain information. CCS Concepts • Human-centered computing → Interactive systems and tools. Keywords Uncertainty management, Large language models, Human-AI interaction 1 INTRODUCTION Large language models (LLMs) are increasingly used to generate long-form answers for knowledge-intensive tasks, such as learning unfamiliar topics, synthesizing background information, drafting analytical reports, and supporting decision-making [23, 87, 111] . In these contexts, users must work through extended responses that may contain factually questionable claims, vague or underspecified explanations, unfamiliar terminology, and statements whose relevance or reliability is difficult to judge [32, 89] . Because these sources of uncertainty are often embedded within otherwise plausible and coherent text, users may struggle to decide which parts of a response deserve scrutiny, why they may be unreliable, and what to do next [18, 46] . Prior work has explored ways of presenting uncertainty cues to users in AI and LLM outputs, including confidence scores, uncertainty markers, visual cues, and natural-language explanations [78, 81, 104]. These approaches are valuable for making uncertainty visible. However, seeing uncertainty is not the same as managing it. A cue can draw attention to possible unreliability, but users still need to locate what the concern i

FAQs

用户在长篇 LLM 回答中的不确定性管理通常包括哪些阶段？

文中将不确定性管理组织为三个阶段：解释（interpretation）、评估（evaluation）和决策（decision）。研究据此分析用户如何处理需要核查的内容并采取下一步行动。

现有工作为什么不足以解决长篇回答中的不确定性管理问题？

以置信度等线索让不确定性可见，并不能等同于管理不确定性。用户仍需定位问题点、理解其重要性并决定是否以及如何行动，尤其在不确定性分散于复杂长文本时更明显。

U-Lens 具体如何支持用户完成不确定性管理？

U-Lens 将不确定性相关信息组织为带上下文的检查目标（inspection targets），对其进行优先级排序，并将每个目标连接到可评估的上下文与可选的后续响应选项，从而覆盖解释、评估与决策过程。

Jul 16, 2026

Upload by Riley