Execution contexts form a key component in the OrdinoR organizational models. This notion enables us to systematically combine multidimensional information in event logs and use that to precisely characterize the involvement of resource groups in process execution. In Chapter 2, we showed examples of case types, activity types, and time types, and how they can be combined to define execution contexts. Those examples can be seen as a result of manually specifying execution contexts based on prior information, such as domain knowledge about an event log and given analysis questions. In this chapter, we introduce an approach that supports automatically learning execution contexts from an event log and explain why it is desirable to have such an approach. Let us revisit the example in Section 2.4. The four activity types imply the existence of an abstract view of the insurance claim process, e.g., “accept claim” and “reject claim” are grouped by type “decide” as they are variants of decisions made on insurance claims, “get missing info” and “pay claim” are grouped by “contact” as both are likely to involve contacting the customer. This abstraction is not directly recorded in the event log (Table 2.1) but may be understood by process owners or analysts who possess relevant domain knowledge. The two time types correspond to a selected level of granularity of timestamps. This categorization of events may be guided by some questions focused on analyzing the performance of human resources during different working hours (morning vs. afternoon). While domain knowledge and guiding questions are key to analyses of event logs, they cannot be assumed to be readily available or sufficiently concrete [80, 38]. Therefore, manually defining execution contexts — as shown in the example — may not always be an option. In the following sections, we will introduce a learning approach that aims at exploiting the discriminative information of events embedded in the data rather than relying on prior information. The approach requires minimal user input and is capable of automatically extracting a set of logic rules from an event log, which can then be used to define high-quality execution contexts.