Everything there is to Know about Stochastically Known Logs

When it comes to data, the world has seen a tremendous change from the days back then when data was neatly stored in well-organized, centrally managed, carefully monitored organizational information systems. Data nowadays can be found in a haphazard collection of datasets, some of which are collected from sensors (humans and artificial) while others are AI generated. Data is available anywhere and everywhere, and its quality is by no means guaranteed or monitored. Process data is no different. What was conceived to be a collection of well-designed protocols, accompanied by traces that record processes in a clear and precise manner, now ranges from noisy labeled data, to partially known processes, through machine tagged activities. The starting point of this paper is the cornerstone of process mining, namely the event data log. We shall inspect the log through the lens of uncertainty, motivating the need for stochastically known logs through modern process mining applications. We shall also investigate the relationships between stochastically known logs and models of probabilistic databases. Then, we will dive into the impact of stochastically known logs on various tasks of process mining. We shall conclude with a discussion of challenges we face as a community when transiting from deterministic logs to stochastically known logs.