【摘要】 大型在线系统在不同终端中的客户端由于兼容问题和频繁迭代容易出现前台显示故障，如控件覆盖、乱码等。由于传统系统后台的指标监控方法无法应对症状繁杂的前台故障，提出利用用户反馈动态检测前台故障的方案，通过对用户反馈的实时分析，挖掘其中关键信息动态构建监控指标，来表征并覆盖各种类型的前台故障。进一步设计快速在海量指标中进行异常检测的两阶段算法，实时地检测出指标中的异常并反映故障。该方法在多个真实大型在线系统中均获得了良好的检测效果，准确率达70%,召回率超过90%。更多还原

【Abstract】 Large-scale online systems have clients applications in different terminal devices with compatibility issues and frequent iteration, which can frequently encounter various front-end issues such as controls cover and messy code. Traditional methods based on monitoring back-end system indicators cannot properly detect such front-end issues. In view of this, we propose a new approach that leverages user feedback text to automatically detect front-end issues. We analyzed the feedback texts in real time, and extracted key information to dynamically construct indicators which could indicate various front-end issues. We further designed a two-stage algorithm to achieve a real-time anomaly detection in those constructed indicators, and in turn detect the occurring front-end issues accordingly. This approach has been applied to several real online service systems to perform front-end issue detection. It proves to be an effective effort, achieving precision at 70% and recall over 90%.更多还原