Historical Data Tacit Knowledge Extraction With LLM Feedback Loops
Find Innovative SolutionsGenerate Solutions
Solution Overview
Problem
Existing systems fail to effectively extract tacit knowledge from historical data, particularly from unstructured sources, and lose domain-specific information over time due to lack of documentation and inefficient information extraction techniques.
Innovation Solution
A method and system that utilize pre-trained large language models (LLMs) to convert large dimensional hyperspaces of tacit knowledge into smaller dimensional hyperspaces by selecting relevant knowledge points based on domain rules, iteratively refining the models to achieve desired output quality, and store the refined models for future use.
Engineering Contradictions & Design Principles
Engineering Contradiction Analysis
1Loss of information
If traditional information extraction techniques are used to extract tacit knowledge from historical data, then the extraction process is simple, but the extracted knowledge loses domain-specific information and tacit knowledge over time
Solution Approach 1:
The patent introduces Large Language Models (LLMs) as intermediary components between historical data and the extraction process. These LLMs act as mediators that understand and preserve domain-specific tacit knowledge while extracting structured information, preventing knowledge loss without requiring complex custom extraction systems
Solution Approach 2:
The system implements iterative feedback loops where extraction results are evaluated against quality thresholds, and domain rules are continuously refined based on performance feedback. This feedback mechanism ensures tacit knowledge is preserved and improved over time while maintaining manageable system complexity
2Loss of information
If all knowledge points from historical data are processed, then complete knowledge extraction is achieved, but computational resources and processing time increase significantly
Solution Approach 1:
The patent extracts and separates domain-specific rules and patterns from historical data into reusable components. By taking out these essential elements, the system processes only relevant knowledge points for each downstream task, achieving complete knowledge extraction without processing unnecessary information, thus reducing time loss
Solution Approach 2:
The system performs preliminary processing to create structured representations of historical data and pre-identify domain rules before actual downstream task execution. This preliminary action organizes knowledge in advance, enabling faster processing during task execution while maintaining completeness
3Measurement precision
If domain rules are continuously refined through iterative processes, then output quality improves, but the processing time and computational effort increase
Solution Approach 1:
The patent implements feedback loops where output quality is evaluated against thresholds, and domain rules are refined iteratively based on performance feedback. This structured feedback mechanism improves output quality systematically while preventing endless refinement cycles, thus maintaining processing efficiency
Solution Approach 2:
The system dynamically adjusts processing parameters and quality thresholds based on task requirements and historical performance. By changing parameters adaptively, the system achieves high output quality when necessary while maintaining efficiency for routine tasks, resolving the contradiction between precision and productivity
Data Source
Figure 1~2
Figure 3
Figure 4A
AI summary
Business rules are currently not documented and are present only as knowledge with subject matter experts (SMEs). The knowledge can be lost with time if it is not extracted or recorded. Existing techniques are unable to extract tacit knowledge and to retain the domain flavor in extracted information. Present disclosure provides a method and a system for extracting tacit knowledge from historical data. The system represents each point in historical data as a large dimensional hyperspace which contains all unstructured information where tacit knowledge can exist. Then, system maps large dimensional hyperspace to smaller dimensional hyperspace using pre-trained large language model (LLM). Thereafter, system, based on the series of downstream tasks, generates a feedback loop to optimally compute dimension of the smaller dimensional hyperspace. Once reduced dimensional space containing effective tacit knowledge information is available, system performs a downstream task based on the extracted tacit knowledge using another pre-trained LLM.