Semantic Story Generation for Predictive Process Monitoring
Find Innovative SolutionsGenerate Solutions
Solution Overview
Problem
Existing predictive process monitoring (PPM) methods, such as deep learning and transformer-based approaches, struggle to effectively capture the semantics of skills and their connections to utterances, especially with sequential data and free-text features.
Innovation Solution
The system transforms event traces into semantic stories using a generated story template and fine-tunes a model with these stories for efficient and accurate task prediction, including next skill prediction.
Engineering Contradictions & Design Principles
Engineering Contradiction Analysis
1Measurement precision
If deep learning or transformer-based methods are used for predictive process monitoring, then prediction capability is improved, but the ability to capture semantics of skills and their connections to utterances deteriorates
Solution Approach 1:
The patent introduces semantic stories as an intermediary representation that bridges raw event traces and prediction models. These stories capture semantic connections between skills and utterances in a structured narrative format, allowing the model to retain semantic information while making predictions. The story generation process acts as a mediator that transforms unstructured event data into meaningful semantic representations.
2Productivity
If traditional machine learning methods are used, then computational efficiency is improved, but the ability to handle sequential data and free-text features deteriorates
Solution Approach 1:
The patent segments the complex task of predicting next skills into multiple components: extracting event traces from event logs, generating semantic stories from traces, and then making predictions. This segmentation allows each component to be optimized independently - trace extraction handles sequential data, story generation processes free-text features, and the prediction model focuses on classification, thereby achieving both efficiency and adaptability.
3Adaptability or versatility
If more automation skills are added to conversational RPA bots, then functionality is improved, but user understanding and selection of appropriate skills deteriorates
Solution Approach 1:
The system uses feedback from historical event logs to generate predictions about which skills users are likely to need next. By analyzing patterns in how users actually interact with the bot and which skills they select in similar situations, the system provides feedback-based recommendations that guide users toward appropriate skills, making the expanded functionality more accessible and easier to operate.
Data Source
AI summary
An example system includes a processor to receive an event trace. The processor can transform the event trace into a semantic story using a generated story template. The processor can input the semantic story into a fine-tuned model. The processor can receive a next skill prediction from the fine-tuned model.