blog

Mining Process Models from Text

Mining Process Models from Text

Accelerating Process Modeling Using NLP

Tracy is a consultant in a digital transformation firm. She just arrived at a new project that requires modeling a large telecommunications company’s processes. Apart from planning a series of meetings with several people in the company, she will need to read a bunch of text documents describing each one of the processes.

This is by no means a bizarre case: many organizations document their processes using plain text. By having a textual description of a process, organizations make sure anyone can understand it. Sometimes having these textual documents is even required by law, like in public administrations, where it is required to show compliance to certain legal principles.

Although having these documents will be very useful to Tracy to understand and model her customer processes, having to read so many of them will take her significant effort and time.
Is technology ready to help Tracy solve her client’s problems faster?

Natural Language Processing (NLP) is a cutting-edge, mature technology for extracting meaning from text. Right from the seventies, the NLP community has provided, during the last forty years, algorithms and tools for automating the processing of human language, enabling today’s AI advanced interaction between humans and machines.

To extract meaning from text, NLP applies a cascade of steps, each one built on the output of the previous: tokenization & sentence splitting, morphological analysis, Part-of-Speech tagging, named entity recognition, syntactic parsing, semantic role labeling, coreference resolution. Until some years ago the majority of these techniques were based on statistical methods, which had reached their limits. During the first decade of the 21st century, GPUs originally designed to speed-up video games started being used to perform fast matrix computations required by artificial neural networks (ANN). Together with the availability of large amounts of data, this led to a big leap in perception tasks such as Computer Vision and Speech processing, and shortly after the technology was applied to more semantic tasks involving language processing: machine translation, summarization, human-machine dialogue systems, etc.
The rebirth of artificial neural network (ANN) technology has made unprecedented progress unthinkable 15 years ago. This has made NLP to be one of the most promising areas for the next decade in industry.

Getting back to Tracy, how can she benefit from these advances? If only there were a system able to read all those documents and produce a representation of the processes described therein in an industry-standard format such as BPMN 2.0…

This is where ProcessTalks makes its appearance!

ProcessTalks uses a stack of NLP technology combining neural and non-neural components to analyze the documents and extract process elements and their relationships. It can detect where a relevant activity or task is mentioned, and who is supposed to perform it. It can detect which are the main actors performing relevant tasks in a process, as well as relationships between activities, such as temporal precedence, or mutually exclusive execution depending on some condition.

For instance, if the text describes part of a process such as “If the marketing department approves the web site, the IT department brings it online. Otherwise, a revision of the content is carried out by the designers, who submit an improved version”, ProcessTalks can generate a representation like:

For each document, Tracy can select the sentences or paragraphs describing the process, and feed them to the process extractor. ProcessTalks will interact with Tracy to display the detected activities and actors, and will allow her to confirm, edit, or remove each element.

Then, a draft BPMN model will be created and opened in ProcessTalks collaborative editor. Tracy can then use ProcessTalks natural language interaction to polish and complete the model. She can also invite her customer to check the model and to interact with it in a simultaneous and collaborative environment.

After a few editing sessions, Tracy produced a model of the company’s processes. She was able to do her job in a much faster way than reading all the documents and hand modeling each process would have required, saving time and money both for her and her customers.

Dare to try? Please schedule a demo with us at this link.

Facebook
Twitter
LinkedIn