Process Instruction Generation from User Logs with Pretrained Language Models
Find Innovative SolutionsGenerate Solutions
Solution Overview
Problem
Existing process instructions do not capture the nuances and variations in how different users perform a process within an organization, often deviating from manually-created instructions due to regulatory changes or more efficient methods, and require resource-intensive neural network training.
Innovation Solution
A computing system uses a trained neural network, based on a foundational large language model, to generate human-readable process instructions by synthesizing data from user interactions, reducing the need for initial pre-training and resource consumption.
Engineering Contradictions & Design Principles
Engineering Contradiction Analysis
1Adaptability or versatility
If a neural network is trained from scratch to generate process instructions, then the model can be customized for specific organizational processes, but the computational resources and time required for training increase significantly
Solution Approach 1:
The patent applies preliminary action by using a pre-trained foundational language model that has already learned general language patterns and knowledge before being applied to the specific task of generating process instructions. This pre-training phase is performed once and reused across multiple organizational contexts, eliminating the need to train from scratch for each new application while maintaining adaptability to specific processes through fine-tuning or prompt engineering.
2Reliability
If manual process documentation is created to capture regulatory compliance, then compliance requirements can be met, but the documentation does not capture variations in how different users perform processes
Solution Approach 1:
The patent applies feedback by analyzing actual user interactions with computing systems to generate process instructions that reflect real-world variations in how different users perform processes. The system continuously learns from logged user actions and uses this feedback to update and refine process instructions, ensuring both regulatory compliance and accuracy in capturing user variations.
Solution Approach 2:
The system performs self-service by automatically generating and updating process instructions based on analyzed user interactions, eliminating the need for manual documentation updates. The machine learning model autonomously captures process variations and generates compliant instructions without human intervention, maintaining both reliability and adaptability.
3Use of energy by stationary object
If existing process instructions are used, then resource consumption is low, but the instructions deviate from actual user behavior and may not comply with current regulations
Solution Approach 1:
The patent applies copying by generating process instructions that are copied or derived from actual user interactions with computing systems. Instead of relying on static existing documentation, the system observes and copies real user behaviors, then synthesizes these observations into accurate process instructions that reflect current practices while maintaining resource efficiency through automated analysis.
Data Source
AI summary
A computing system may access log files that log data associated with prior performances of a process by users. Such data may be recorded by one or more computing systems based on tracking user interactions with the one or more computing systems to perform the process. The computing system may generate, using an instructions generation model that is trained using machine learning, the process instructions for performing the process based on the data associated with the prior performances of the process. The process instructions may be human-readable written instructions.