Entity Information Graph Construction Using Fusion and Link Prediction
Find Innovative SolutionsGenerate Solutions
Solution Overview
Problem
Existing information graph construction methods lack a unified representation system, leading to inaccurate reasoning and incomplete information due to complex relationships between entities, resulting in ambiguous and incomplete information graphs.
Innovation Solution
A method involving information extraction, fusion, and link prediction to generate entity information graphs, using scoring and loss functions to ensure accuracy and completeness, including structuring data, performing word segmentation and classification, and updating the initial graph based on link prediction.
Engineering Contradictions & Design Principles
Engineering Contradiction Analysis
1Measurement precision
If traditional information graph construction methods are used, then the construction process is simple, but the reasoning accuracy is low and information completeness is poor
Solution Approach 1:
The information graph construction process is divided into distinct stages: information extraction to generate triples, information fusion to integrate multiple sources, and link prediction to infer relationships. This segmentation allows each stage to be optimized independently, improving overall reasoning accuracy while managing complexity through structured progression.
Solution Approach 2:
Before constructing the final information graph, the method performs preliminary information extraction to generate structured triples, followed by information fusion to integrate data from multiple sources. These preliminary actions ensure high-quality input data for the link prediction stage, thereby improving reasoning accuracy before the actual graph construction occurs.
2Loss of information
If traditional information graph construction methods are used, then the processing speed is fast, but the information completeness is poor
Solution Approach 1:
The method merges information from multiple sources through the information fusion stage, combining triples generated from different information extraction processes. This integration reduces information loss by consolidating data from diverse sources into a comprehensive information graph, ensuring more complete representation of real-world entities and relationships.
Solution Approach 2:
The link prediction stage uses the constructed information graph to infer new relationships, and these inferred relationships feed back into the graph structure. This feedback mechanism continuously improves information completeness by adding previously unknown relationships while maintaining construction efficiency through automated inference processes.
3Reliability
If traditional information graph construction methods are used, then the construction process is simple, but the logical relationships are insufficient
Solution Approach 1:
The method introduces structured triples as an intermediary representation between raw information and the final information graph. These triples serve as a standardized intermediate format that captures logical relationships explicitly, enabling more accurate reasoning while managing complexity through a clear transformation pipeline from extraction to fusion to graph construction.
4Manufacturing precision
If unstructured information data is directly used, then the data processing is simple, but the structured representation is poor
Solution Approach 1:
The information extraction process segments unstructured data into discrete triples with defined components (subject, predicate, object). This segmentation transforms complex unstructured information into structured units that can be systematically processed and integrated, improving data representation accuracy while managing processing complexity through modular extraction rules.
Data Source
AI summary
The present disclosure relates to information graph generation technology, and discloses a method, an electronic apparatus, and a storage medium for generating an entity information graph. The method includes: acquiring entity information data, and performing information extraction on the entity information data to obtain a triplet; performing information fusion on the triplet to obtain fused information, and constructing a scoring function and a loss function according to the fused information; constructing an initial information graph based on the scoring function and the loss function using the fused information, and performing link prediction on the initial information graph; and updating the initial information graph based on a result of the link prediction to obtain an entity information graph. The present disclosure can mine valuable information data from massive information sources, convert unstructured information data into structured data, and generate an information graph with logical relationships based on the structured data.