Document Segment Tagging With GNN-RNN for Faster Receipt Parsing
Find Innovative SolutionsGenerate Solutions
Solution Overview
Problem
Current methods for extracting purchase data from purchase documents are time-consuming, resource-intensive, prone to errors, and costly, and traditional AI models for segment tagging are computationally expensive and inefficient, especially when dealing with unstructured documents like receipts.
Innovation Solution
Employing a graph neural network (GNN) and recurrent neural network (RNN) combination for segment tagging, using graph attention networks (GAN) to enrich node embeddings with neighbor information and incorporate sequential data, reducing computational resources and improving accuracy.
Engineering Contradictions & Design Principles
Engineering Contradiction Analysis
1Measurement precision
If traditional AI models are used for segment tagging, then comprehensive feature analysis is achieved, but computational resources and processing time increase significantly
Solution Approach 1:
The patent segments the document processing task into distinct components: graph construction from text segments, GNN-based structural feature extraction, and RNN-based sequential feature extraction. This segmentation allows each component to be optimized independently, achieving accurate segment tagging without the computational overhead of monolithic traditional AI models.
Solution Approach 2:
The patent introduces graph structures as an intermediary representation between raw text segments and final tagging results. The GNN processes this intermediate graph representation to extract structural features, which are then combined with sequential features from RNN. This intermediary approach enables efficient feature extraction while maintaining high tagging accuracy.
2Measurement precision
If traditional AI models are used for segment tagging, then accurate predictions are achieved, but computational resources and cost increase
Solution Approach 1:
The GNN processes the graph structure constructed from text segments itself, extracting structural features directly from the data representation. The RNN similarly processes the sequential information self-sufficiently. This self-service approach eliminates the need for complex traditional AI model architectures, reducing computational resource requirements while maintaining prediction accuracy.
Solution Approach 2:
The patent changes the parameter representation by using graph-based structural features and sequential features instead of traditional high-dimensional features. This parameter transformation enables more efficient computation while preserving the information necessary for accurate segment tagging predictions.
3Measurement precision
If comprehensive feature extraction is performed, then tagging accuracy improves, but processing complexity increases
Solution Approach 1:
The patent segments feature extraction into two distinct pathways: structural features from GNN processing of graph data structures, and sequential features from RNN processing of text sequences. This segmentation simplifies the overall model complexity by breaking down comprehensive feature extraction into manageable, specialized components rather than using a single complex model.
Solution Approach 2:
The graph structure serves multiple functions: it represents the spatial relationships between text segments, enables structural feature extraction via GNN, and provides a foundation for combining with sequential features. This multi-functionality achieves comprehensive feature extraction without proportionally increasing model complexity.
Data Source
AI summary
Methods, apparatus, systems, and articles of manufacture are disclosed to tag segments in a document. An example apparatus includes processor circuitry to execute machine readable instructions to generate node embeddings for nodes of a graph, the node embeddings based on features extracted from text segments detected in a document, the text segments to be represented by the nodes of the graph; sample edges corresponding to the nodes to generate the graph; generate first updated node embeddings by passing the node embeddings and the graph through layers of a graph neural network, the first updated embeddings corresponding to the node embeddings augmented with neighbor information; generate second updated node embeddings by passing the first updated embeddings through layers of a recurrent neural network, the second updated embeddings corresponding to the first updated node embeddings augmented with sequential information; and classify the text segments based on the second updated node embeddings.