Knowledge Graph Claim Processing With OCR and LLM Inference
Find Innovative SolutionsGenerate Solutions
Solution Overview
Problem
The insurance industry faces inefficiencies in processing structured and unstructured text documents, particularly handwritten text, leading to lengthy claim adjudication cycles, increased resource costs, and computational complexity, especially with foreign-language claims, which negatively impact employee satisfaction and customer satisfaction.
Innovation Solution
Utilizing specialized graph data structures in conjunction with machine learning computing architectures to process and analyze insurance claims, including optical character recognition (OCR) and large language models (LLMs) to convert handwritten text to digitized format, populate knowledge graphs with key healthcare and demographic details, and generate personalized recommendations for claimants.
Engineering Contradictions & Design Principles
Engineering Contradiction Analysis
1Measurement precision
If traditional rule-based computing is used to process handwritten text documents, then processing accuracy can be maintained through explicit rules, but computational complexity becomes extremely high and processing time exceeds 30 days
Solution Approach 1:
The patent replaces traditional mechanical rule-based computing systems with machine learning-based systems. Specifically, it uses neural networks and natural language processing models to automatically extract and structure information from handwritten documents, eliminating the need for complex explicit rule engines while maintaining or improving processing accuracy.
Solution Approach 2:
The patent transforms the processing approach by changing parameters from deterministic rule evaluation to probabilistic machine learning predictions. This includes using confidence scores, probability distributions, and adaptive thresholding to handle unstructured data, fundamentally altering how processing decisions are made and reducing computational complexity.
2Reliability
If manual claims adjudication is performed by specialists, then accurate decision-making can be achieved, but processing duration exceeds 30 days and resource costs increase
Solution Approach 1:
The patent introduces machine learning models as intermediary systems between raw document data and final adjudication decisions. These models pre-process and structure unstructured handwritten data into standardized formats, enabling automated processing while maintaining decision quality through human-in-the-loop validation for complex cases.
Solution Approach 2:
The patent segments the claims processing workflow into distinct automated and human-reviewed portions. Simple, clear-cut claims are fully automated through machine learning extraction and decision-making, while only ambiguous or complex cases require specialist review, thereby dramatically increasing overall productivity without sacrificing adjudication quality.
3Speed
If claims are approved without prejudice to meet short cycle times, then processing speed increases, but unrecoverable costs occur if claims are later declined
Solution Approach 1:
The patent performs preliminary structuring and analysis of handwritten documents using machine learning models before final adjudication decisions are made. This pre-processing extracts key entities, relationships, and risk indicators early in the workflow, enabling faster processing while maintaining accuracy through informed decision-making based on structured data rather than rushed approvals.
4Measurement precision
If foreign-language claims are translated before processing, then processing accuracy can be maintained, but cycle time increases by weeks
Solution Approach 1:
The patent merges multiple processing functions into a single integrated machine learning pipeline that simultaneously performs optical character recognition, language identification, translation, and information extraction. This consolidation eliminates sequential processing steps and their associated delays, maintaining accuracy while dramatically reducing the time loss from translation operations.
Data Source
AI summary
A system and method for knowledge graph data structure based machine learning capable of processing claim documents to reduce cycle time and improve client experience and return to work outcomes is proposed. The system uses a Large Language Model (LLM), having a name entity recognition model and next best action engine, to identify claim specific data. The claim specific data may include keywords and phrases which are related to a user's claim or history. A pre-trained knowledge graph data structure trained based on a corpus of similar historical treatments is provided to the next best action engine in combination with the user's information for operating in inference mode. The nest best action engine contains triggers which correspond to the identified entities or relationships, and automatically generates an output data structure.