Aligning textual and model-based process descriptions

Abstract

Process model descriptions are an ubiquitous source of information that exists in any organization. To reach different types of stakeholders, distinct descriptions are often kept, so that process understandability is boosted with respect to individual capabilities. While the use of distinct representations allows more stakeholders to interpret process information, it also poses a considerable challenge: to keep different process descriptions aligned. In this paper, a novel technique to align process models and textual descriptions is proposed. The technique is grounded on projecting knowledge extracted from these two representations into a uniform representation that is amenable for comparison. It applies a tailored linguistic analysis of each description, so that the important information is considered when aligning description' elements. Compared to existing approaches that address this use case, our technique provides more comprehensive alignments, which encompass process model activities, events, and gateways. Furthermore, the technique, which has has been implemented into the platform nlp4bpm.cs.upc.edu, shows promising results based on experiments with real-world data.

Supporting the Process of Learning and Teaching Process Models

“…The Model Judge Exercise Editor: the text describing the process (right) can be annotated (left) and relations between annotations reported (arcs in the text). [12].…”

“…The goal of aligning a process model and a textual process description is to identify correspondences between the elements of the process model and parts of the textual process description [12], [33], [34]. In particular, all elements that Type Text Fragment ↵ 1 action "an examination and follow-up treatment order is placed by the physician" ↵ 2 action "is then responsible for taking a sample to be analysed in the lab" ↵ 3 action "After receiving the sample, a physician of the lab validates its state" ↵ 4 role "an examination and follow-up treatment order is placed by the physician" ↵ 5 role "the female patient is examined by an outpatient physician" ↵ 6 object "an examination and follow-up treatment order is placed by the physician" ↵ 7…”

“…In this paper we extend our alignment framework presented in earlier works [12], [34] by assuming that textual descriptions are annotated. This assumption allows strengthening some of the constraints in the algorithm by having the certainty that information that comes from annotated texts is precise and complete.…”

“…The framework proposed is inspired by similar practices in other educational contexts, e.g., the success of judges in supporting learning to program [7]- [11]. Remarkably, we use natural language processing (NLP), together with optimization techniques [12], to validate process models with respect to textual descriptions of processes.…”

“…Model Judge provides diagnostics regarding issues on syntactic, pragmatic and semantic quality. The framework underlying Model Judge is grounded on aligning annotated textual descriptions with the models as basis for providing diagnostics [12]. Using Model Judge, novice modelers can be guided through continuous feedback.…”