Document Structure Extraction via ML Tree Vectors
Find Innovative SolutionsGenerate Solutions
Solution Overview
Problem
Existing methods for analyzing data in documents ignore structural information, leading to inefficiency and waste of computational resources, as they consider elements in isolation without accounting for document hierarchy or relationships.
Innovation Solution
A method using machine learning to extract structural information by constructing trees from document elements, encoding node values into vectors, sampling pairs of vectors through random walks, and training a second machine learning model to learn common document structures, allowing for efficient analysis and downstream tasks like data scraping.
Engineering Contradictions & Design Principles
Engineering Contradiction Analysis
1Productivity
If existing classification techniques analyze document elements in isolation, then the analysis process is simple, but the computational efficiency is low and structural information is lost
Solution Approach 1:
The patent segments the document analysis process into multiple stages: first constructing a tree structure to represent document hierarchy, then encoding node values into vectors, sampling vector pairs through random walks, and finally training a machine learning model. This segmentation allows the system to process structural information efficiently without analyzing entire documents at once, resolving the contradiction between computational efficiency and analysis complexity.
Solution Approach 2:
The patent performs preliminary actions by pre-processing document elements into tree structures and pre-computing node encodings before the main analysis task. The random walk sampling and vector pair generation are performed in advance to create training data, which reduces the computational burden during actual analysis and improves overall productivity.
2Loss of information
If language models analyze documents as a whole, then the model can capture global context, but the computational resources are wasted and processing time increases
Solution Approach 1:
The patent extracts structural information from documents by building tree representations that capture hierarchical relationships. Instead of analyzing entire documents, the system extracts relevant node values and their structural positions, encoding them into vectors. This extraction approach retains essential structural information while avoiding the computational overhead of processing complete documents, thus reducing processing time.
Solution Approach 2:
The patent transforms document structure into a different dimension by representing hierarchical relationships as tree structures and converting node values into vector representations. The random walk sampling operates in this vector space, allowing the model to capture structural relationships without processing the full document text, thereby reducing processing time while maintaining structural information.
3Measurement precision
If existing methods ignore document structure, then the analysis is faster and simpler, but the accuracy of information extraction decreases
Solution Approach 1:
The patent applies local quality by encoding structural information at the node level rather than treating the entire document uniformly. Each node in the tree structure receives a localized encoding based on its position and content, allowing the model to focus computational resources on relevant local structures. This approach improves extraction accuracy by capturing local hierarchical relationships while maintaining processing efficiency.
Solution Approach 2:
The patent changes parameters by transforming document elements into vector representations with specific dimensions that encode both content and structural information. The random walk sampling generates vector pairs with specific statistical properties that capture hierarchical relationships. These parameter transformations enable the model to achieve high extraction accuracy by representing structural information in a form suitable for machine learning processing.
Data Source
AI summary
The present disclosure provides techniques for extracting structural information using machine learning. One example method includes receiving electronic data indicating one or more pages, constructing, for each page of the one or more pages, a tree based on the page, wherein each level of the tree includes one or more nodes corresponding to elements in a level of elements in the page, encoding, for each page of the one or more pages, a value of each node of the tree for the page into a vector using a first machine learning model, sampling a plurality of pairs of vectors from the one or more trees for the one or more pages, wherein a given pair of vectors corresponds to values of nodes in a same tree, training a second machine learning model using the plurality of pairs, and combining each vector with weights of the second machine learning model.