paperbot · PL 论文追踪

全部论文 待读 收藏

P2 PosEyeDOM: Lightweight Positional Copilot Suggestion Logger for Eye-Tracking Applications using the DOM

PosEyeDOM:面向眼动追踪应用的基于 DOM 的轻量级 Copilot 建议位置记录器
ICSE Proceedings of the 2026 IEEE/ACM 48th International Conference on Software Engineering2026匹配度 2/10 · 该论文聚焦于用Chrome扩展记录Copilot建议的位置与时机以支持眼动追踪研究,属于HCI/开发者工具方向,不涉及静态分析、形式化验证、浮点程序分析或贝叶斯程序分析。虽然与AI辅助编程相关,但其研究目标并非PL与LLM/Agent的交叉分析,匹配度较低。引用 0 inbox
Tarek Alakmeh, Sarah D’Angelo, Thomas Fritz
crossref: 2026-09-12

AI 速览

一句话:PosEyeDOM:基于 DOM 的轻量级 Chrome 扩展,实时记录 Copilot 建议的位置、内容与时机,推断接受结果并导出眼动研究数据。

理解 GenAI 代码建议对开发者生产力的影响,需要研究开发者与内联建议的真实交互。然而,GitHub Copilot 等助手的建议出现和消失极快、随滚动移动,且从单行补全到多行浮层形态各异;现有工具对未被采纳即消失的瞬态覆盖层支持有限,眼动追踪又依赖精确的位置与时序数据。PosEyeDOM 以一个轻量级 Chrome 扩展填补这一空白,面向 GitHub Codespaces 等基于 Web 的 VS Code 环境,利用 DOM 实时记录建议的几何信息、内容与时机,推断接受结果,并聚合为稳健且带标签的兴趣区(AOI)。它可导出兼容 SR Research DataViewer 的 .ias 文件及供自定义流程使用的 JSON 字典,并提供实时预览、远程导出及与外部眼动软件同步,使以往不可行的、可扩展的建议感知眼动研究成为可能。

问题
内联 GenAI 代码建议出现/消失快、随滚动移动且形态多样,眼动追踪难以精确捕捉其位置与时机;现有工具对未被采纳即消失的瞬态建议浮层支持有限。
方法
PosEyeDOM 以轻量级 Chrome 扩展形式,借助 DOM 在 Web VS Code 环境中实时记录建议的几何、内容与时机,推断接受结果,并聚合为带标签的兴趣区(AOI)。
评估
通过导出兼容 SR Research DataViewer 的 .ias 文件与 JSON 字典,并提供实时预览、远程导出和与外部眼动软件同步的即插即用流程,支持可扩展的建议感知眼动研究。
原文摘要(Abstract)

Understanding the impact of GenAI code suggestions on developer productivity requires studying how developers actually interact with them. Yet, inline suggestions from assistants such as GitHub Copilot appear and disappear quickly, shift with scrolling, and range from single-line completions to multi-line overlays. Capturing interactions is especially difficult for eye-tracking, which depends on precise positional data and timing. Existing tools offer limited support, as these transient overlays may disappear without ever being integrated into the source code. PosEyeDOM addresses this gap with a lightweight Chrome extension for web-based VS Code environments such as GitHub Codespaces. By leveraging the Document Object Model (DOM), it continuously logs suggestion geometry, content, and timing in real-time, infers acceptance outcomes, and aggregates them into robust and labeled Areas of Interest (AOIs). It exports ready-to-use .ias files compatible with SR Research DataViewer and JSON dictionaries for custom pipelines, enabling scalable, suggestion-aware eye-tracking studies that were previously impractical. The extension further provides a live preview, remote export, and synchronization with external eye-tracking software, offering an all-in-one, plug-and-play workflow for empirical research on AI-assisted programming.

链接与引用

DOI 原文 · PDF(开放获取) ·

BibTeX
@inproceedings{paperbot11680,
  title = {PosEyeDOM: Lightweight Positional Copilot Suggestion Logger for Eye-Tracking Applications using the DOM},
  author = {Tarek Alakmeh and Sarah D’Angelo and Thomas Fritz},
  booktitle = {ICSE},
  year = {2026},
  doi = {10.1145/3774748.3787608}
}