Machine-Learning Document Graphs for Semantic Clause Search
Find Innovative SolutionsGenerate Solutions
Solution Overview
Problem
Conventional document management systems lack the ability to support complex agreement processes, such as locating relevant clauses, tracking entities involved in negotiations, and determining how agreements have been amended over time, due to limitations in lexical searching and lack of semantic understanding.
Innovation Solution
A document graph structure is implemented, utilizing machine learning models to transform unstructured text into structured nodes and edges, enabling semantic similarity searches to find semantically similar information within a graph database, allowing for accurate retrieval of relevant document portions and relationships.
Engineering Contradictions & Design Principles
Engineering Contradiction Analysis
1Ease of manufacture
If lexical searching is used to locate information in electronic documents, then the search process is simple to implement, but the search accuracy and ability to understand semantic meaning deteriorates
Solution Approach 1:
The patent replaces traditional lexical search mechanisms with machine learning-based semantic search. The system uses trained ML models to transform search queries and document text into vector embeddings, enabling semantic similarity searches that understand meaning rather than just matching keywords. This substitution resolves the contradiction by achieving both implementation feasibility through automated ML pipelines and superior search accuracy through semantic understanding.
Solution Approach 2:
The patent transforms the search approach by changing parameters from exact keyword matching to semantic similarity measurement. By converting text into vector representations and measuring cosine similarity or other distance metrics, the system achieves more accurate results while maintaining implementability through standardized ML operations. This parameter transformation resolves the contradiction between simple implementation and accurate results.
2Device complexity
If conventional document management systems are used to track agreements and amendments, then the system structure is simple, but the ability to locate relevant clauses and track entities deteriorates
Solution Approach 1:
The patent enhances the document management system by adding a semantic dimension through vector embeddings and graph structures. Instead of relying solely on traditional hierarchical document structures, the system creates semantic relationships between clauses, entities, and amendments using graph databases. This dimensional addition improves information retrieval capability while maintaining manageable system complexity through modular architecture.
Solution Approach 2:
The patent introduces machine learning models and graph database structures as intermediaries between the document storage system and the query interface. These intermediaries transform unstructured document text into structured semantic representations, enabling efficient tracking of entities and amendments. The intermediary layer resolves the contradiction by improving information retrieval without requiring complete redesign of the underlying system structure.
3Productivity
If existing agreement generation systems query agreement storage databases, then the systems can retrieve information, but the queries return incomplete data or incorrect information
Solution Approach 1:
The patent implements feedback mechanisms where the system continuously learns from query results and improves its semantic understanding. The ML models are trained on retrieved data and refined based on accuracy metrics, creating a feedback loop that simultaneously maintains fast retrieval speeds through optimized vector search while improving data accuracy through iterative model training. This feedback mechanism resolves the contradiction between productivity and reliability.
Solution Approach 2:
The patent performs preliminary actions by pre-processing and transforming document text into vector embeddings before queries are executed. The system pre-structures the knowledge base using graph databases and trains ML models in advance, enabling fast and accurate query execution without compromising data quality. This preliminary preparation resolves the contradiction by ensuring both speed and accuracy are achieved before the actual retrieval operation.
Data Source
AI summary
A method, an apparatus, and a computer-readable storage medium for generating a document graph. A plurality of electronic documents is received. Each electronic document has a predetermined document type. A machine learning model is selected from the plurality of machine learning models based on the predetermined document type. The selected machine learning model is instructed to extract a plurality of document portions from each electronic document in the plurality of electronic documents in accordance with the predetermined document type. A relationship between two or more document portions is defined based on the content of each document portion, and the document portions are associated based on the relationship. A graph structure having a plurality of nodes is generated. Each node includes at least one document portion. Each node is connected to another node in accordance with the relationship between document portions included in the nodes. The graph structure is stored.