Information Extraction Using Entity-Attribute Graph Correlation
Find Innovative SolutionsGenerate Solutions
Solution Overview
Problem
Conventional information extraction methods face challenges such as semantic drifting and sparse data issues, where irrelevant entities are extracted and attributes are incorrectly identified due to separate entity and attribute extraction processes.
Innovation Solution
An information extraction method that establishes correlations between entities and attributes using an entity-attribute graph, ranking entities and attributes based on correlation coefficients to select target entities and attributes, thereby integrating the extraction process and addressing semantic drifting and sparse data issues.
Engineering Contradictions & Design Principles
Engineering Contradiction Analysis
1Ease of manufacture
If separate entity and attribute extraction processes are used, then the extraction process is simpler to implement, but semantic drifting occurs and irrelevant entities are extracted
Solution Approach 1:
The patent merges the previously separate entity extraction and attribute extraction processes into a unified joint extraction framework. The entity-attribute graph integrates both extraction tasks, allowing them to be performed simultaneously with shared computational resources and coordinated optimization, thereby improving extraction accuracy while maintaining implementation feasibility.
Solution Approach 2:
The patent introduces an entity-attribute graph as an intermediary structure that connects entities and attributes. This graph serves as a mediator that facilitates coordinated extraction by establishing relationships between entities and their attributes, enabling the system to leverage contextual information and reduce semantic drifting without requiring completely new extraction mechanisms.
2Device complexity
If separate entity and attribute extraction processes are used, then the processing pipeline is more straightforward, but attributes are incorrectly identified due to sparse data issues
Solution Approach 1:
By combining entity and attribute extraction into a unified process through the entity-attribute graph, the patent enables shared computation and information flow between the two tasks. This joint processing allows the system to leverage entity context for attribute identification and vice versa, improving attribute identification accuracy even with sparse data, while the overall pipeline complexity remains manageable due to the structured graph approach.
3Reliability
If joint entity-attribute extraction is performed, then extraction performance and accuracy are enhanced, but the computational complexity increases
Solution Approach 1:
The patent segments the joint extraction problem into manageable components through the entity-attribute graph structure. The graph divides entities and attributes into discrete nodes with defined relationships, allowing the computational task to be broken down into localized operations on graph edges and nodes. This segmentation reduces overall computational complexity while maintaining the benefits of joint extraction.
Solution Approach 2:
The patent applies local quality by focusing computational resources on specific entity-attribute relationships within the graph rather than processing all possible combinations globally. The entity-attribute graph enables localized extraction operations that concentrate computation on relevant pairs, improving efficiency and reducing overall computational complexity while maintaining high extraction accuracy.
Data Source
AI summary
The present application discloses an information extraction method. The method includes extracting a plurality of candidate entities and a plurality of candidate attributes from a target corpus based on a plurality of entity seeds of one or more target types in an entity seeds set; establishing a plurality of correlations between the plurality of candidate attributes and a plurality of entities including the plurality of candidate entities and the plurality of entity seeds; and selecting one or more target entities from the plurality of entities and one or more target attributes from the plurality of candidate attributes based on the plurality of correlations.