LLMs-For-Subject-Oriented-ProcessModeling

This repository contains the source code, prompts, and generated visualizations for Masters Thesis on:
Exploring the Potential of LLM Technology to Create Subject-Oriented Process Models from Natural Language Texts.
It includes implementations for zero-shot, one-shot, and chain-of-thought (CoT) prompting strategies, as well as a web-based interactive Graphical User Interface (GUI) for interactive human-in-the-loop PASS process modeling.

Repository Structure

This repository has the following structure:

LLMs-For-Subject-Oriented-ProcessModeling/  
├── Outputs/  
│   ├── Thesis_Illustrations/               # Full-size vector graphics used in the thesis  
│   ├── CoT/                                # CoT visualizations  
│   ├── OWL/                                # OWL visualizations  
│   ├── OneShotPrompting/                   # OneShotPrompting visualizations  
│   └── ZeroShotPrompting/                  # ZeroShotPrompting visualizations  
├── src/                                    # Source code (all scripts)
    ├── CoT/
    ├── GUI/                                # GUI notebooks
        ├── GUI-Interactive/                # Final Web-based human-in-the-loop interactive GUI notebooks
        ├── Llama/                          # Original Llama notebook (non-interactive)
        ├── Mistral-Small/                  # Original Mistral notebook (non-interactive)
    ├── OWL/ 
    ├── OneShotPrompting/
    ├── ZeroShotPrompting/ 
├── README.md                               # This file  
└── Requirements.txt                        # Python dependencies

Model-Specific Folders

• Llama: Contains code used for experiments with Llama model. Experiments are organized by prompting strategy: zero-shot, one-shot, and chain-of-thought.

• Mistral-Small: Contains code used for experiments with Mistral-Small model. Experiments are organized by prompting strategy: zero-shot, one-shot, and chain-of-thought.

The variable model in each notebook determines which model is used in the notebook. You can change it to your preferred model (e.g., "mistral-small").

Installation

Clone the repository and install dependencies:

git clone https://github.com/shadiamanan/LLMs-For-Subject-Oriented-ProcessModeling.git  
cd LLMs-For-Subject-Oriented-ProcessModeling  
pip install -r Requirements.txt

Configuration

Before running the notebooks, configure your API key and preferred model.

from openai import OpenAI

api_key = "YOUR_API_KEY"
model = "Llama-3.3-70B"  # change to your preferred model
client = OpenAI(api_key=api_key, base_url=base_url)

Model Usage Notes

Mistral: Best for linear simple scenarios.
Responses may be truncated if the context window is exceeded.

Llama: Recommended for complex scenarios requiring decision flows and longer context window.

File Paths

The notebooks reference local directories for saving diagrams or files (e.g., E:\Thesis\PASS Diagrams\Diagram\llama).

To run the notebooks, you can either:

1. Create a folder structure on your local machine where the outputs will be saved, or
2. Update the path variables in the notebooks to match a directory of your choice.

Running the GUI in Jupyter Notebook

The repository provides a web-based, human-in-the-loop interactive GUI for exploring different LLMs in process modeling.

Steps to launch

Each GUI notebook has a last cell that launches the web-based interactive GUI.

• The top of this cell contains the model variable, which determines which LLM will be used in the GUI:

model = "mistral-small"  # You can change it to your preferred model  

Open the corresponding GUI notebook and run the last cell in the notebook. It ends with:

app.show()

the web-based interactive GUI will launch.

Model Details for GUI

• Llama folder: contains code using Llama model. This GUI uses one-shot complex scenario prompt.
• Mistral-Small folder: contains code using Mistral-Small model. This GUI uses one-shot simple scenario prompt.

Prompts used differ between the two GUI versions.

The variable model in each notebook determines which model is used in the GUI. You can change it to your preferred model (e.g., "mistral-small"). All GUI calls will use this variable.

GUI Notebook Notes

• The final GUI notebooks in the GUI-Interactive folder use the one-shot prompting strategy, which produced the best overall results in experiments.

Configuring for a Different LLM

If you are not using the Uni Münster LLM, update the following variables:

base_url variable in the notebook to match your LLM provider’s endpoint. For example:

base_url = "https://api.openai.com/v1"  # OpenAI endpoint

model variable in the notebook for using LLM of choice. For example:

model = "mistral-small"  # model used  

In GUI notebooks, the model and base_url variables are at the top of the last cell in the notebook. In other notebooks, update the variables in initialization cells for a specific prompt strategy.