Business Process Management is facing a drift in the way process information is used within an organization. To reach a wide audience, organizations keep parallel representations of process information, thus making the processes to be understandable by everyone. However, this poses a challenge on the synchronization and transformation between di↵erent process representations. NLP4BPM is an environment to support such crucial tasks. It combines Natural Language Processing tools with process-oriented techniques, and through its web-based interface can be easily accessed from any device.
Process documentation is a key aspect of Business Process Management.
Nowadays organizations store processes descriptions in various representations. The
reason for this is the di↵erent nature stakeholders have: while textual descriptions
of processes are well-suited for non-technical members, they are less appropriate
for describing precise aspects of the underlying process [
In the last decade, the field of Natural Language Processing (NLP) has
become mature enough to provide text analysis accuracies accurate enough for
a variety of applications. Currently, there are several powerful open-source
libraries [
There are several approaches one might consider in order to automate dealing with consistency issues in multiple representations. One approach is to generate a textual description in natural language using the most up-to-date process model. There is also the complementary technique, converting the textual description into a formal process model. The third approach consists on performing a comparison of two existing process representations.
In this paper, we introduce NLP4BPM, an online platform o↵ering NLP
based services for the aforementioned BPM use-cases: Converting between
several process model representations as well as comparing process models between
themselves and to textual descriptions. The interface presented covers
adaptations or modifications of techniques that have appeared in the last years [
The tool we present o↵ers several functionalities in the context of semantic
understanding of business processes, all of which use natural language processing
techniques:
Text2BPMN: Converts a textual description into a BPMN process model.
BPMN2Text: Converts a BPMN process model into a textual description.
BPMNvsText: Computes an alignment between a model and a textual
description in order to compare them. An alingment is a mapping between
tasks in the model and sentences in the text. The implementation follows
the technique presented in [
The di↵erent techniques are developed as independent modules and unified under the same graphical interface, which is presented in a user-friendly manner. A screencast of the tool can be found at https://youtu.be/mAsXnEBQMx0 2.1
An overview of the architecture of the project can be seen in Figure 1. The core of the application are the four main modules, implemented as separate Java projects. The graphical interface is build as a JavaServer Faces (JSF) web application which integrates the four main modules, and provides a global configuration interface to all of them. By using the JSF framework, we obtain a robust way to integrate Java code into a responsive web application that can be accessed from all kinds of devices: From personal computers to smartphones.
At the very core of all four techniques, there is natural language
processing. In order to o↵er a scalable, robust system, we forward the NLP tasks
to TextServer 1, a distributed REST-like web service running FreeLing [
The tool is currently under heavy development with the four main modules being improved. The comparison modules: BPMNvsBPMN and BPMNvsText are currently at a stable state, while the translation ones: BPMN2Text and Text2BPMN are early prototypes still being improved. 2.2
The NLP4BPM web application is available from nlp4bpm.cs.upc.edu. After login, a main menu is shown to the user with four tabs corresponding to each of the described modules, as seen in Figure 2. The left menu contains the configuration parameters of each algorithm. The configurations vary for each module, but all share a set of common parameters: Language Due to FreeLing’s multi-language support, our tool also provides support for performing tasks on many languages besides English, among which we can find Spanish, French or German. Note that this feature is under construction for some of the modules.
Input Files The supported input files are BPMN XML files for the process models and raw text files for the textual descriptions. Additional format support is planned as a future extension for several rich document formats as well as di↵erent process model notations. Also, the support for the (recently IEEE standard) XES event logs format will be added soon. The users can upload their own input files or use one of the provided examples.
The four modules o↵er a very similar interface. Once the user has set the module-specific parameters and specified the textual description and/or BPMN model, the Execute button runs the corresponding algorithm. 1 textserver.cs.upc.edu/textserver 2 nlp.cs.upc.edu/freeling
In the translation modules, the user is shown the resulting BPMN model or textual description, which is saved in the user’s online directory and can be downloaded. In the comparison models, a graphical representation of the comparison is shown to the user.
Figure 3 shows an example of the resulting comparison from the BPMNvsText module. A BPMN process model is shown with its tasks and gateways highlighted in colors. The colors code is used on the sentences below, in order to show the alignment. For additional information, the user can click one of the matchings to get information about which linguistic features were found in common for the sentence and the task.
By providing an easy-to-use environment for the transformation and comparison of process descriptions, organizations can reach an unprecedented process awareness. This may be crucial to avoid wrong expectations on the processes within a company, but also to ensure all the important agents running these processes are aligned.
We also foresee interesting applications of the environment proposed. For
instance, the environment can be used to assist in the process of process
modeling [