| Property | Value |
|---|---|
| Parameter Count | 108M |
| Base Model | BERT-base-cased |
| License | Apache 2.0 |
| F1 Score | 90.31% |
| Accuracy | 95.16% |
What is bpmn-information-extraction-v2?
This is a specialized token classification model built on BERT-base-cased architecture, designed to extract structured information from business process descriptions. The model has been fine-tuned on a dataset of 104 textual process descriptions and can identify five key elements: Agents, Tasks, Task Information, Process Information, and Conditions.
Implementation Details
The model leverages a fine-tuned BERT architecture with 108M parameters, trained using the Adam optimizer with a learning rate of 2e-05 over 15 epochs. The training process utilized a batch size of 8 and achieved impressive metrics with 88.26% precision and 92.46% recall.
- Token Classification Architecture
- PyTorch Implementation
- TensorBoard Integration
- Safetensors Support
Core Capabilities
- Extraction of process agents and actors
- Identification of business tasks and activities
- Recognition of conditional statements in process flows
- Classification of process-related metadata
- Structured information extraction from natural language descriptions
Frequently Asked Questions
Q: What makes this model unique?
The model's specialization in business process text analysis and its high accuracy (95.16%) in identifying process elements make it particularly valuable for automated business process modeling and analysis.
Q: What are the recommended use cases?
The model is ideal for converting natural language process descriptions into structured BPMN elements, automating business process documentation, and analyzing workflow descriptions for process mining and optimization.