arXivDaily arXiv每日学术速递 周一至周五更新
arXiv 2210.09569cs.HC
收藏

ModSandbox:通过自动化规则的错误预测与改进促进在线社区审核

ModSandbox: Facilitating Online Community Moderation Through Error Prediction and Improvement of Automated Rules

Jean Y. Song, Sangwook Lee, Jisoo Lee, Mina Kim, Juho Kim

AI总结:

本文针对 AutoModerator 规则难以预判误报、漏报及更新的问题,提出可视化虚拟沙盒 ModSandbox,帮助审核员定位规则缺陷并改进。

摘要:

尽管基于规则的工具被普遍用于在线内容审核,但人工审核员仍需花费大量时间监控这些工具,以确保其按预期运行。基于对使用 AutoModerator 的 Reddit 审核员的问卷调查和访谈,我们确定了减少自动化规则误报和漏报的主要挑战:无法提前评估规则的实际效果,并且难以弄清应如何更新规则。为解决这些问题,我们构建了 ModSandbox,这是一种新颖的虚拟沙盒系统,可检测待改进规则可能存在的误报和漏报,并可视化规则中导致问题的具体部分。我们对在线内容审核员开展了用户研究,发现 ModSandbox 能够支持快速发现自动化规则可能存在的误报和漏报,并指导审核员更新这些规则,以减少未来的错误。

英文摘要:

Despite the common use of rule-based tools for online content moderation, human moderators still spend a lot of time monitoring them to ensure that they work as intended. Based on surveys and interviews with Reddit moderators who use AutoModerator, we identified the main challenges in reducing false positives and false negatives of automated rules: not being able to estimate the actual effect of a rule in advance and having difficulty figuring out how the rules should be updated. To address these issues, we built ModSandbox, a novel virtual sandbox system that detects possible false positives and false negatives of a rule to be improved and visualizes which part of the rule is causing issues. We conducted a user study with online content moderators, finding that ModSandbox can support quickly finding possible false positives and false negatives of automated rules and guide moderators to update those to reduce future errors.

机构、总结、中文摘要均由AI分析而来,可能存在错误,请以论文正文为准。

↑