Computer Modeling System for Automated Subrogation Detection
Find Innovative SolutionsGenerate Solutions
Solution Overview
Problem
Insurance companies frequently miss subrogation opportunities due to the manual and inefficient process of claims adjusters detecting potential subrogation, leading to lost chances of recovering losses from third parties.
Innovation Solution
A computer modeling system that processes unstructured textual data from claims logs to identify potential subrogation opportunities by training models on subsets of data categorized by loss types, using natural language processing techniques and Support Vector Machine classification to predict subrogation likelihood, thereby automating the detection of qualifying claims.
Engineering Contradictions & Design Principles
Engineering Contradiction Analysis
1Measurement precision
If manual review by claims adjusters is used to identify subrogation opportunities, then human judgment and context understanding are applied, but the process is inefficient and time-consuming with minutes per claim
Solution Approach 1:
The patent replaces the manual mechanical review process by claims adjusters with an automated computer modeling system that uses natural language processing and machine learning algorithms to analyze claims data, thereby maintaining detection accuracy while dramatically increasing processing speed
Solution Approach 2:
The patent introduces trained classification models as intermediaries between raw claims data and subrogation identification decisions. These models process unstructured textual data through multiple stages including vocabulary extraction, feature engineering, and classification to bridge the gap between manual review accuracy and automated efficiency
2Reliability
If manual review processes are used by claims adjusters, then flexible judgment is applied, but human error and inconsistency lead to missed subrogation opportunities
Solution Approach 1:
The patent segments the subrogation detection system into multiple specialized models trained on different loss types (water, fire, vehicle damage). This segmentation allows each model to specialize in specific claim categories, improving reliability and consistency while managing complexity through modular design
Solution Approach 2:
The patent transforms unstructured textual claims data into structured features through natural language processing techniques including vocabulary extraction, bigram formation, and feature engineering. This parameter transformation enables consistent quantitative analysis while maintaining the ability to capture nuanced information from unstructured text
3Productivity
If automated computer modeling is implemented to process claims data, then processing speed increases to tens of thousands of claims per minute, but the complexity of data processing and model training increases
Solution Approach 1:
The patent divides the complex processing task into distinct stages: vocabulary extraction from unstructured text, bigram generation, feature selection, and classification. Multiple specialized models are trained on different loss type subsets, managing complexity through systematic segmentation of both data and processing steps
Solution Approach 2:
The patent performs preliminary processing of claims data including vocabulary extraction, bigram formation, and feature engineering before classification. This preliminary action prepares the data in advance, enabling faster processing during actual subrogation detection while managing complexity through pre-computed features
Data Source
AI summary
Improved computer modeling techniques, and uses thereof, are described herein. A set of unstructured textual data is received. Certain textual data is removed from the set of unstructured data to form an initial vocabulary set of textual data. One or more bigrams are added to the vocabulary set of textual data to form a final vocabulary set of textual data. The final vocabulary set of textual data is divided into a plurality of subsets of textual data based on type. A model is trained using each of the plurality of subsets of textual data to form a plurality of trained models, each corresponding to one of the types.