论文标题
ModSandBox：通过错误预测和自动化规则改进来促进在线社区审核
ModSandbox: Facilitating Online Community Moderation Through Error Prediction and Improvement of Automated Rules
论文作者
论文摘要
尽管通常将基于规则的工具用于在线内容审核中，但人类主持人仍然花费大量时间监视它们以确保其按预期工作。根据对使用自动调节器的Reddit主持人进行的调查和访谈，我们确定了减少误报和虚假自动化规则的错误挑战：无法提前估算规则的实际效果并难以确定应该如何更新规则。为了解决这些问题，我们构建了ModSandBox，这是一种新颖的虚拟沙盒系统，它检测出可能改善规则的误报和假否定性，并可视化规则的哪一部分引起了问题。我们通过在线内容主持人进行了一项用户研究，发现ModSandBox可以支持快速找到可能的误报和自动化规则的假否定性，并指导主持人更新这些规则以减少未来错误。
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.
