Virtual Knowledge Graph Rule Mining from Text
Find Innovative SolutionsGenerate Solutions
Solution Overview
Problem
The construction of knowledge graphs from text data is a time-consuming and expensive process, especially for large amounts of unstructured or domain-specific texts, making it impractical for applications like machine reasoning and natural language processing.
Innovation Solution
An unsupervised method for constructing virtual knowledge graphs from text by generating named entity pairs with relation embedding vectors, clustering these vectors to form a relation-cluster graph, and extracting logical rules for predictive models.
Engineering Contradictions & Design Principles
Engineering Contradiction Analysis
1Measurement precision
If knowledge graphs are constructed manually or semi-automatically from structured sources or text with human annotations, then the quality and accuracy of the knowledge graph is improved, but the time consumption and cost increase significantly
Solution Approach 1:
The system performs self-service by automatically constructing knowledge graphs from unstructured text without requiring human annotations or manual intervention. The unsupervised learning algorithm autonomously extracts entities, relations, and constructs the knowledge graph structure, eliminating the need for human labor while maintaining reasonable quality.
Solution Approach 2:
The patent replaces the mechanical human annotation process with an automated computational system. Instead of manual extraction and construction, the system uses unsupervised learning algorithms, embedding models, and automated relation extraction to build knowledge graphs, substituting human mechanical work with computational processes.
2Measurement precision
If knowledge graphs are constructed manually or semi-automatically, then the accuracy of extracted logical rules is improved, but the productivity and scalability decrease
Solution Approach 1:
The system changes the parameters of the construction process by using unsupervised learning instead of supervised learning, adjusting the approach from requiring labeled data to working with unstructured text. This parameter change enables processing large volumes of text automatically while extracting meaningful logical rules through relation clustering and path analysis.
Solution Approach 2:
The system creates virtual knowledge graphs as computational copies that represent the structure and relationships from the text data. These virtual KGs can be processed, analyzed, and used for rule extraction without requiring physical manual construction, enabling rapid iteration and scalability.
3Quantity of substance
If large amounts of unstructured or domain-specific texts are processed, then the quantity of extracted information increases, but the complexity and cost of knowledge graph construction becomes impractical
Solution Approach 1:
The system segments the complex task of knowledge graph construction into distinct components: entity extraction, relation extraction, embedding generation, relation clustering, and rule mining. This segmentation allows each component to be processed independently and automatically, managing the complexity of handling large amounts of unstructured text.
Solution Approach 2:
The patent introduces embedding vectors as an intermediary representation between the raw text and the knowledge graph structure. These embeddings capture semantic relationships and enable automated relation clustering, serving as a mediator that simplifies the construction process from unstructured text to structured knowledge representations.
Data Source
AI summary
A method, system, and computer program product are disclosed. The method includes receiving an input text and generating a set of virtual triples, which include pairs of named entities from the input text and relation embedding vectors for each of the pairs, from the input text. The method also includes constructing a virtual knowledge graph (KG) with the set of virtual triples and transforming the virtual KG into a relation-cluster KG. Further, the method includes mining logical rules from the relation-cluster KG.