Synthetic Chemical Document Generation for AI Training Consistency
Find Innovative SolutionsGenerate Solutions
Solution Overview
Problem
Existing methods for generating training datasets for AI models to recognize molecular structures in chemical documents are labor-intensive, prone to errors, and result in inconsistent data formats, limited appearance styles, and high costs, making them inefficient and costly.
Innovation Solution
A computer-implemented method for generating synthetic data sets of chemical documents using configuration parameters, structure, and content generation modules to create complex layouts with chemical objects, noise objects, and diverse styles, allowing for automated and efficient dataset creation.
Engineering Contradictions & Design Principles
Engineering Contradiction Analysis
1Manufacturing precision
If manual annotation is used to create training datasets, then dataset quality and consistency can be maintained, but labor intensity and development time increase significantly
Solution Approach 1:
The patent uses template-based synthetic document generation that copies and adapts real document structures and styles to create training data. Templates capture the essential layout patterns, chemical illustration arrangements, and formatting conventions of authentic chemical documents, allowing automated generation of realistic training samples without manual annotation of each document
Solution Approach 2:
The patent performs preliminary actions by pre-defining document templates with proper structures, styles, and formatting rules before dataset generation. These templates encapsulate the consistency requirements and layout conventions that would otherwise need to be manually enforced during annotation, enabling automated generation of consistent training data
2Reliability
If manual annotation is used to create training datasets, then specialized domain knowledge can be incorporated, but costs and development time increase
Solution Approach 1:
The patent copies authentic chemical document characteristics including domain-specific layouts, chemical illustration styles, and formatting conventions into reusable templates. This allows automated generation of domain-appropriate training data without requiring manual annotation experts for each document while preserving the reliability of domain-specific appearances
Solution Approach 2:
The patent creates universal templates that can generate multiple types of chemical documents (research papers, patent applications, technical reports) with a single system. The templates are designed to accommodate various document structures and chemical illustration types, reducing the need for separate annotation processes for different document categories
3Stability of the object's composition
If fixed appearance styles are used in manually created datasets, then consistency is maintained, but adaptability to different document types is limited
Solution Approach 1:
The patent implements dynamic template selection and configuration that adapts to different document types and requirements. The system can dynamically choose appropriate templates based on document category, adjust layout parameters, and modify chemical illustration arrangements to match the specific style requirements of different chemical document types while maintaining internal consistency within each generated document
Solution Approach 2:
The patent segments the document generation process into independent template components (layout structure, chemical illustration placement, formatting styles) that can be independently selected and configured. This allows the system to assemble different combinations of components to create diverse document types while maintaining consistency through the modular template structure
4Productivity
If automated methods are used to generate synthetic datasets, then productivity increases, but manufacturing precision and realism may deteriorate
Solution Approach 1:
The patent uses template-based synthetic document generation that copies and adapts real document structures and styles to create training data. Templates capture the essential layout patterns, chemical illustration arrangements, and formatting conventions of authentic chemical documents, allowing automated generation of realistic training samples without manual annotation
Solution Approach 2:
The patent employs configurable template parameters that can be adjusted to match specific document requirements. By changing parameters such as layout dimensions, chemical illustration sizes, spacing, and formatting options, the system generates diverse yet realistic documents while maintaining structural accuracy through the underlying template framework
Data Source
AI summary
The invention is notably directed to a computer-implemented method to generate a synthetic data set. The synthetic data set comprises a plurality of chemical documents and each of the plurality of chemical documents comprises a respective set of chemical objects. The method comprises a step of receiving configuration data, the configuration data comprising a set of configuration parameters for the data set. The method further comprises performing, in an iterative manner, for each of the plurality of chemical documents the steps of generating, by a structure generation module, a respective document structure for each of the respective chemical documents in accordance with the configuration data, generating, by a content generation module, the respective set of chemical objects for the respective chemical document and arranging the respective set of chemical objects on the respective chemical document.