Trie-Based Topic Modeling for Low-Context Document Classification
Find Innovative SolutionsGenerate Solutions
Solution Overview
Problem
Existing AI models struggle to accurately categorize documents with little meaningful contextual information, such as those in key-value pairs or tabular formats, due to limited training data and inconsistent formats, leading to degraded performance.
Innovation Solution
A novel trie structure is used to categorize unstructured data by generating a dictionary from frequent base words in historical document images, forming N-gram groups, and creating a trie structure for efficient classification.
Engineering Contradictions & Design Principles
Engineering Contradiction Analysis
1Measurement precision
If a large amount of high quality training data is provided to improve model accuracy, then topic prediction accuracy improves, but data availability deteriorates due to privacy and confidentiality constraints
Solution Approach 1:
The patent introduces an intermediary processing layer that transforms private, structured document data into synthetic training data through text generation. This intermediary process allows the system to create training datasets without directly exposing or transferring the original confidential information, thereby resolving the contradiction between needing large training datasets and data privacy constraints
Solution Approach 2:
The system creates synthetic copies of training data by generating text that mimics the patterns and structures of private documents. These synthetic copies serve as substitutes for the original confidential data, enabling model training while preserving the privacy and security of the source information
2Reliability
If traditional ML models are used to detect intent and theme from text, then contextual relationship analysis improves, but performance deteriorates on documents with little meaningful contextual information
Solution Approach 1:
The patent fundamentally changes the input parameters for topic detection by shifting from analyzing contextual relationships in original text to analyzing frequency distributions of base words in generated text. This parameter transformation allows the system to effectively process documents with minimal contextual information, such as key-value pairs and tabular data, while maintaining reliable topic detection performance
Solution Approach 2:
The system segments the document processing into distinct stages: extracting base words from structured data, generating text that preserves frequency relationships, and then analyzing the generated text for topic detection. This segmentation allows each stage to be optimized independently, improving overall reliability across diverse document formats
3Measurement precision
If comprehensive text analysis is performed to improve categorization accuracy, then classification precision improves, but computational intensity increases
Solution Approach 1:
The patent extracts only the essential frequency distribution patterns from documents by identifying and counting base words, rather than performing comprehensive text analysis. This extraction approach captures the critical information needed for accurate categorization while significantly reducing computational intensity compared to traditional NLP methods
Data Source
AI summary
Method includes obtaining historical document images including text that correspond to different document classes; and generating a dictionary using text of the historical document images. The dictionary includes base words occurring with a greatest frequency in each document class. The base words are extracted from the text of the historical document images and arranged in datasets by a document class, where each dataset includes the base words of a same document class that occur with the greatest frequency within that document class. Trie structure is generated using the base words of the datasets that occur with a greatest frequency in each dataset. The trie structure includes internal nodes including root node and leaf nodes in which keys corresponding to the base words occurring with the greatest frequency in each dataset are respectively stored in predefined order. The trie structure is searchable in the predefined order starting with the root node.