Testing Representational Biases

A set of 120 event logs was created to test the effect of the representational bias used and the effect of erratic and infrequent behavior added to highly regular behavior. For each log the underlying model is known. Therefore, such as “gold standard model” serves as a reference for the real process. For real-life event logs there is no such a reference model. Moreover, quality criteria like precision and generalization are still subject to discussion. Therefore, these synthetic event logs were created. The hope is that these event logs contribute to better process discovery tools that can deal with different representational biases and different types of noisy behavior.本研究构建了120条事件日志（event logs），用于评估所采用的表征偏差（representational bias）的影响，以及为高度规整行为添加异常且低频行为后的影响。每条日志对应的底层模型均为已知，此类模型即所谓"金标准模型（gold standard model）"，可作为真实业务流程的参照基准。而现实场景中的事件日志并无此类参照模型，此外，诸如精确度（precision）与泛化性（generalization）等质量评估标准仍存在争议。正因如此，本研究构建了这批合成事件日志。本研究期望此类事件日志能够助力开发出可适配不同表征偏差与各类噪声行为的更优质流程发现工具。