Process Model Node Label Recommendation via Fine-Tuned Language Model
Find Innovative SolutionsGenerate Solutions
Solution Overview
Problem
Establishing accurate models of complex process flows is time-consuming and error-prone due to reliance on domain experts unfamiliar with model creation, and existing repository models limit label recommendations to existing vocabularies, restricting flexibility and accuracy.
Innovation Solution
A fine-tuned sequence-to-sequence language model is generated using transfer-learning techniques from a pre-trained model like T5, allowing for node label recommendations beyond the limited vocabulary of repository models, enhancing the accuracy and flexibility of modeling complex process flows by extending the vocabulary and recommending appropriate labels for incomplete process models.
Engineering Contradictions & Design Principles
Engineering Contradiction Analysis
1Reliability
If domain experts are used to create process models, then the models reflect accurate domain knowledge, but the modeling process becomes time-consuming and error-prone due to their unfamiliarity with model creation
Solution Approach 1:
An AI assistant acts as an intermediary between domain experts and process modeling tools. The assistant translates natural language domain knowledge into formal process models, eliminating the need for experts to learn modeling syntax while preserving the accuracy of their domain knowledge.
Solution Approach 2:
The system enables automated self-service modeling by allowing domain knowledge to be directly transformed into process models through AI, without requiring manual intervention from experts in the model creation process themselves.
2Stability of the object's composition
If repository models with existing vocabularies are used for label recommendations, then consistency within the repository is maintained, but flexibility and accuracy are restricted to the limited vocabulary
Solution Approach 1:
The label recommendation system dynamically adapts to new terminology by learning from the context of process models. Instead of being constrained by a static vocabulary, the system can recommend labels with new terms while maintaining consistency with the overall process model structure and semantics.
Solution Approach 2:
The system changes the parameter of vocabulary flexibility by allowing the recommendation engine to suggest labels outside the existing repository vocabulary when contextually appropriate, thus adapting to domain-specific terminology needs while maintaining model consistency.
3Adaptability or versatility
If a pre-trained language model is used, then general language understanding is available, but specific process modeling accuracy is insufficient without fine-tuning
Solution Approach 1:
The system performs preliminary fine-tuning of the language model on process modeling data before deployment. This preliminary action ensures that when the model is used for actual label recommendations, it has both general language understanding and specific process modeling accuracy.
Data Source
AI summary
Embodiments afford recommendations in the accurate modeling of complex process flows. A repository is provided of known models (in graph form) of complex processes. Semantics of the repository models, are constrained within an existing vocabulary (e.g., one that does not include a particular term). During an initial training phase, a fine-tuned sequence-to-sequence language model is generated from a pre-trained language model (e.g., T5) and semantics of the known repository process models, using transfer-learning techniques (e.g., from Natural Language Processing—NLP). During runtime, an incomplete process model (also in graph form) is received having an unlabeled node. Embodiments provide a node label recommendation based upon the fine-tuned sequence-to-sequence language model. The node label that is recommended, is in a vocabulary which extends beyond the repository vocabulary (e.g., includes the particular term). In this manner, accuracy and/or flexibility of modeling of complex processes (e.g., node label recommendation) can be enhanced.