Process models are a fundamental element in the BPM lifecycle. Hence, it is of paramount importance for organizations to rely on high-quality, accurate and up-to-date process models, to avoid taking decisions on the basis of a wrong picture of the reality. In this demo we present modeljudge.cs.upc.edu, a platform to boost the training of novice modelers when confronted with the task of translating a textual description into a process model in BPMN notation. The platform is integrated with Natural Language Processing (NLP) analysis and textual annotation, together with a novel model-to-text alignment technique. By using this platform, a novice modeler will receive diagnostics in real-time, which may contribute to a more satisfactory modeling experience.
Due to the wide usage of process models in organizations, correctness and quality of models have a direct influence in the execution of business processes. However, research has shown that industrial process models often contain errors, which can lead to many problems, like increased costs in production.
Automating the detection of syntactic errors is a common feature in modeling
software. However, the error types more closely related to the natural language sections
of the model are usually not checked, due to the difficulties in the automatic analysis
of such elements. Model Judge is a web platform supporting students in the creation
of business process models by automatically detecting and reporting the most common
sources of semantic and pragmatic errors in modeling. The algorithm for the
computation of diagnostics is based on the technique for automatic computation of alignments
between process model and textual descriptions presented in [
Significance of the tool for the BPM field. As it is pointed out in [
The Model Judge is presented as a web-based platform, that can be accessed through any web browser at http://modeljudge.cs.upc.edu. It is designed both for helping students in the process of creating a process model and instructors in the task of designing modeling activities in an agile way.
Working with the Model Judge is very much like creating a process model using any BPMN editor. What makes the platform special is the underlying engine to provide automatic feedback (see below). When modeling, a textual description of the process model is available on the left part, while the modeling editor is shown in the right part. Fig. 1 shows the workspace.
With the goal of providing an accurate evaluation of students’ models, and inspired
by the success of judges to support learning to program ([
Two different types of feedback can be provided to the student, depending on the granularity of the information required: – Validation: returns an aggregated diagnostic that reflects if the model has some errors of the types explained before, but it does not say what or where is exactly
the problem. The motivation for this check is to allow for a mild test to guide the students without giving away the whole solution. The feedback provided in Fig. 2 is a validation. – Complete Validation: apart from the overall information provided by the Validation, this check also provides a detailed list of all the problems detected, and individually explained. The motivation for this check is to allow an assessment similar to the one obtained if a teacher was correcting the model, and would be typically provided once the student finishes and hands over the exercise.
In order to use the platform, a user (either student or instructor) needs an account, with its associated storage space. This space acts as a cloud drive for models, allowing to manually save multiple versions of an exercise. Moreover, the students enrolled in a particular course, may enable the platform to record a history of their modeling session. This can be used for analyzing their behavior, which can be reported back to the instructor to get a clearer picture of the modeling process of their students.
Currently, there are 9 different exercises available in the platform. Any modeler can register into the platform and practice with them. Instructors can also design a course, by selecting the exercises that must be included. Fig. 3 shows the main page for defining a course as instructor. Once a course is created, a unique code will be created that can be shared with the students of this course.
Support for adding new exercises is restricted to the developers of the platform. However, support is planned to allow instructors to create their own exercises. In order to create a new exercise for the Model Judge, a textual description of a process is required. This textual description will be used as the problem statement for the students so they can understand the process to be modeled. Instructors will then have to annotate the relevant parts of the process: Actions, Entities and Conditions as well as their relations: Agent, Patient. This annotation process is partially performed by a Natural Language Processing algorithm, which provides an initial annotation to be refined. Figure 4 shows a fragment of a text annotation corresponding to one of the exercises in the platform.
A screencast of the Model Judge which shows a typical session working with the model judge can be found in https://youtu.be/xJ3TeKlvIfo.
Model Judge is built as a web application. A distributed server manages several
instances of the application and balances the load between them. The front-end of it is
built using the PrimeFaces Java framework, which internally communicates with the
core Java application, responsible for the generation of diagnostics. Several
functionalities in Model Judge rely on external libraries: FreeLing [
The Model Judge has been tested in two separate modeling courses. The first was performed on the Technical University of Denmark (DTU) during February 2018. The second course was performed at the Catholic University of Santa Mar´ıa (UCSM) in Peru during March 2018. For every student of these courses, we stored periodically (every minute) information for the whole modeling session. Additionally, information was also saved each time the user performed a simple or complete validation. In particular, we recorded a total of 8410 intermediate models for 72 students.