ML Action Item Extraction from Construction Manuals
Find Innovative SolutionsGenerate Solutions
Solution Overview
Problem
Current computerized systems are inadequate for accurately extracting and categorizing construction action items from construction project manuals, as they often rely on simple parsing or copying, which is not feasible due to the natural language and lack of formatting standards in these documents, resulting in low accuracy and inefficiency.
Innovation Solution
The implementation of a self-learning system that employs a positive-unlabeled learning classification algorithm and a multi-class classification algorithm to automatically extract and categorize action items from construction project manuals, using Optical Character Recognition (OCR) and pattern matching to refine extraction and assignment to appropriate workflows, with a graphical user interface for user confirmation and feedback.
Engineering Contradictions & Design Principles
Engineering Contradiction Analysis
1Measurement precision
If simple parsing or copying methods are used to extract action items from construction manuals, then the process is simple and fast, but the accuracy is low due to natural language formatting variations
Solution Approach 1:
The system employs self-learning machine learning models that automatically improve their extraction accuracy through positive-unlabeled learning. The models learn from the data itself without requiring manual annotation of every example, enabling the system to serve itself and continuously improve performance while maintaining reasonable complexity
Solution Approach 2:
The system implements feedback mechanisms where extraction results are evaluated and used to retrain the machine learning models. This closed-loop feedback allows the system to learn from its mistakes and improve accuracy over time, resolving the contradiction between accuracy and complexity
2Measurement precision
If machine learning models are used to extract and categorize action items, then accuracy is significantly improved, but computational resources and processing time increase
Solution Approach 1:
The system performs preliminary actions by pre-processing the construction manual text and preparing training data before the actual extraction task. The machine learning models are trained in advance on representative datasets, so when actual extraction is needed, the models are already optimized and ready to process documents efficiently
Solution Approach 2:
The system dynamically adjusts its processing based on the specific document being analyzed. The machine learning models can adapt their behavior based on the characteristics of the input document, optimizing the balance between accuracy and processing time for different types of construction manuals
3Measurement precision
If manual review and correction of extracted action items is performed, then accuracy is improved, but productivity and efficiency decrease
Solution Approach 1:
The system performs self-correction through automated feedback loops where extraction results are evaluated and used to retrain the machine learning models. This self-service capability reduces or eliminates the need for manual review while maintaining high accuracy, thereby preserving productivity
Solution Approach 2:
Automated feedback mechanisms evaluate extraction results and automatically retrain models to improve accuracy. This closed-loop system replaces manual review processes with automated self-improvement, maintaining high accuracy without sacrificing processing throughput
4Adaptability or versatility
If the system continuously learns from user feedback, then it can self-improve over time, but system complexity and computational requirements increase
Solution Approach 1:
The system implements self-service learning where the machine learning models automatically incorporate user feedback and retrain themselves without requiring complex external training infrastructure. This self-directed learning approach enables continuous improvement while keeping the system architecture relatively simple
Solution Approach 2:
The system uses feedback from user corrections to automatically retrain the machine learning models. This feedback-driven learning mechanism enables the system to adapt and improve over time while maintaining a straightforward implementation approach
Data Source
AI summary
Systems, methods, and other embodiments associated with automatic smart extraction and workflow assignment of action items are described. In one embodiment, a method includes extracting a set of candidate action items from text of a construction project manual; applying static rules to the candidate action items to distinguish valid and invalid candidate action items; evaluating each valid candidate action item with a first machine learning model to label the valid candidate action item either (i) a true action item or (ii) not a true action item; evaluating each true action item with a second machine learning model to allocate each of the true action items to a construction workflow class; and transmitting the set of true action items to a submittal exchange system to populate one or more workflows.