Data-driven process analyses revolve around event data and the behavior of the entities involved in the process that are represented therein. To derive meaningful insights, views on the event data are required. However, while still in an exploratory stage, an analyst must explore potential views to identify those that are suitable for the intended analysis. To avoid the materialization of all possible views for exploration, which would lead to data redundancy and increased complexity, we introduced the view materialization problem for context exploration in process analysis. First, we define contexts as groups of events with a relation representing the behavior of the captured entities. Both, the set of contexts and the relation, together constitute a view. The view materialization problem then aims to find those k views that best cover the behavior represented in the event data. We instantiated the problem for object-centric event data, leveraging techniques from the literature to induce object-driven contexts, and showed how to solve it based on techniques for subset selection. A prototypical implementation demonstrates the feasibility of the approach for synthetic and real-world data.
Basmer et al. (Thu,) studied this question.