IP Library › Granted Patent US 12,608,372

Granted Patent B2

US 12,608,372 · App. 18/597,955 · Granted Apr 21, 2026

System and method for automated analysis of legal documents within and across specific fields

Inventors: Jason Crabtree (Vienna, VA); Andrew Sellers (Monument, CO)

Assignee: QOMPLX LLC

G06F16/245 G06F16/248 G06F16/9024 G06F40/30 G06N5/04

View Patent ↗

Loading inventors, assignments & file history…

Monitor This Case

Get email alerts when status or documents change.

Order Certified Copies

Most orders are placed with the USPTO same day — all within 24 business hours.

Order via The Patent Place →

Pre-filled with this patent's details

Quick Facts

Patent No.: US 12,608,372
App. No.: 18/597,955
Granted: Apr 21, 2026
Kind: B2

External Links

USPTO Patent Center↗ Full Patent PDF↗

Abstract

Automated analysis of legal documents within and across different fields is disclosed. An extraction processor identifies and extracts knowledge from data contained in documents and transforms it into a common data form. An analysis processor develops local and global knowledge graphs containing the key entities, relationships and concepts encoded in the text.

Claims (90)

1 . A computer system for automated analysis of legal documents within and across different fields employing a cyber decision platform, wherein the computer system is configured to execute software instructions stored on a nontransitory machine-readable storage media, the computer system comprising:

one or more hardware processors configured for:

receiving a data set containing documents including legal documents, technical documents, and domain-specific documents from multiple jurisdictions and practice areas;

hierarchically classifying documents as legal or domain-specific to other fields including engineering, technology, cybersecurity and medicine, wherein legal documents are routed to legal-specific processing and non-legal documents are routed to domain-specific specialty models;

automatically classifying the legal documents by legal domain, jurisdictional framework, and temporal context using machine learning classification algorithms;

performing dynamic model selection of one or more models from a curated model catalogue of domain-specific natural language processing (NLP) models based on the classification, wherein the domain-specific NLP models are trained for specific specialties and jurisdictional requirements;

extracting legal entities including laws, regulations, cases, judgments, contracts, legal concepts, parties, jurisdictions, organizations, people, events, locations, and dates, contractual clauses, and precedential relationships from the received legal documents using the selected domain-specific NLP models;

generating jurisdiction-specific knowledge graphs that link legal concepts, case precedents, regulatory requirements, and contractual provisions with weighted relationship scores, wherein the knowledge graphs comprise:

nodes representing legal entities, contractual terms, statutory provisions, and case precedents;

edges representing legal relationships weighted by precedential strength and jurisdictional applicability; and

temporal annotations tracking legal concept evolution and regulatory changes over time;

performing real-time knowledge graph enrichment using current legal developments, regulatory changes, and litigation outcomes; and

providing predictive legal risk assessments and outcome probabilities for the received legal documents by:

querying the jurisdiction-specific knowledge graphs using the extracted legal entities and relationships from the received documents;

analyzing relevant case precedents and regulatory requirements identified in the knowledge graphs; and

generating document-specific risk predictions based on historical litigation data and current legal trends integrated within the knowledge graphs.

2 . The computer system of claim 1 , wherein dynamic model selection comprises:

analyzing document metadata to determine legal practice area from a predefined taxonomy of legal specialties;

identifying jurisdictional markers including court references, statutory citations, and regulatory framework indicators;

selecting specialized legal NLP models trained on domain-specific legal corpora; and

applying jurisdiction-specific legal ontologies and terminologies to normalize extracted legal concepts.

3 . The computer system of claim 1 , wherein the jurisdiction-specific knowledge graphs comprise:

nodes representing legal entities, contractual terms, statutory provisions, and case precedents;

edges representing legal relationships weighted by precedential strength and jurisdictional applicability; and

temporal annotations tracking legal concept evolution and regulatory changes over time.

4 . The computer system of claim 1 , wherein generating jurisdiction specific-knowledge graphs further comprises:

applying reinforcement learning algorithms to optimize knowledge graph entity classification and relationship placement based on prediction accuracy feedback from the predictive legal risk assessments; and

continuously refining knowledge graph structure through iterative learning processes that improve legal entity relationship accuracy over time.

5 . The computer system of claim 4 , wherein the reinforcement learning algorithms incorporate human feedback from legal experts, users, and crowd-sourced annotations to improve knowledge graph structure and legal entity relationships through reinforcement learning from human feedback (RLHF).

6 . The computer system of claim 1 , further comprising:

generating system recommendations for relevant cases, expert content, legal decisions, findings, and research papers using reinforcement learning algorithms trained on user interaction patterns and expert feedback;

applying reinforcement learning from human feedback to refine link suggestions and relationship determinations within the knowledge graphs; and

providing contextual recommendations based on the jurisdiction-specific knowledge graphs and current legal document analysis.

7 . The computer system of claim 1 , wherein the jurisdiction-specific knowledge graphs include temporal enhancements comprising:

time-weighted relationship scores that decay based on legal precedent age and current relevance;

temporal clustering of legal concepts to track regulatory evolution patterns;

dynamic updating of legal relationship strengths based on recent case outcomes and regulatory changes;

historical versioning of legal knowledge graph states to enable temporal legal analysis.

8 . The computer system of claim 1 , wherein hierarchical classification further comprises:

accessing specialty dictionaries for non-legal domains including engineering, technology, cybersecurity, medicine, and other technical specialties;

applying domain-specific terminologies and ontologies for each identified specialty field;

routing cross-domain documents to hybrid processing pipelines that combine legal and technical domain expertise; and

generating cross-domain knowledge graphs that link legal concepts with technical domain concepts for comprehensive analysis.

9 . The computer system of claim 1 , wherein knowledge graph enrichment comprises:

implementing hierarchical extraction and analysis processes analogous to GraphRAG approaches when processing both a priori reference materials and new user query inputs;

generating community summaries of related legal concepts analogous to local graph clusters for focused analysis;

maintaining both local knowledge graphs for document-specific analysis and global knowledge graphs for comprehensive legal domain coverage;

applying lazy evaluation techniques for efficient knowledge graph querying and dynamic expansion based on query requirements; and

integrating community detection algorithms to identify legal concept clusters and relationship patterns within the jurisdiction-specific knowledge graphs.

10 . The computer system of claim 1 , wherein extracting legal entities further comprises:

identifying and extracting core legal entities including:

laws and regulations with hierarchical structure including sections and subsections;

cases and judgments including parties involved, rulings, and associated legal documents;

contracts including terms, clauses, and obligations;

legal concepts including abstract legal principles and doctrines;

parties including individuals, organizations, and entities involved in legal matters;

jurisdictions including geographical areas where legal rules apply;

establishing relationships between entities including “applies to,” “cited in,” “involved in,” “governs,” and “is an example of” relationships; and

weighting relationships based on legal precedential strength, jurisdictional authority, and temporal relevance.

11 . The computer system of claim 1 , wherein providing predictive legal risk assessments further comprises:

analyzing legal brief arguments and cited case law to determine relative strengths and weaknesses of legal positions;

performing fact pattern comparison between current documents and historical case outcomes;

tracking judicial decision patterns and preferences for outcome probability modeling;

identifying optimal legal strategies based on historical success rates in similar jurisdictional contexts; and

generating confidence intervals for legal outcome predictions based on knowledge graph relationship strengths and historical litigation data patterns.

12 . A computer-implemented method executed on a cyber decision platform for automated analysis of legal documents within and across different fields, the computer-implemented method comprising:

receiving a data set containing documents including legal documents, technical documents, and domain-specific documents from multiple jurisdictions and practice areas;

hierarchically classifying documents as legal or domain-specific to other fields including engineering, technology, cybersecurity, and medicine, wherein legal documents are routed to legal-specific processing and non-legal documents are routed to domain-specific specialty models;

automatically classifying the legal documents by legal domain, jurisdictional framework, and temporal context using machine learning classification algorithms;

performing dynamic model selection of one or more models from a curated model catalogue of domain-specific NLP models based on the classification, wherein the domain-specific NLP models are trained for specific legal specialties and jurisdictional requirements;