Bootstrapping Machine for Relation Extraction
Find Innovative SolutionsGenerate Solutions
Solution Overview
Problem
Reliable relation extraction in text analysis is challenging due to varying wording and descriptions of relationships, which affects the accuracy and reliability of extracting relevant n-tuples from natural language documents.
Innovation Solution
A method and system for relation extraction using seeds, including template and entity seeds, that employs a similarity metric for iterative bootstrapping, where instances are extracted and clustered based on reliability, improving the hit rate and reliability through successive hops, and utilizing positive and negative seeds to refine the extraction process.
Engineering Contradictions & Design Principles
Engineering Contradiction Analysis
1Reliability
If traditional relation extraction methods are used, then the extraction process is simple, but the reliability and accuracy of extracted instances are low due to varying wording and descriptions
Solution Approach 1:
The patent segments the relation extraction process into multiple successive hops, where each hop extracts instances based on seeds from the previous hop. This multi-stage approach improves reliability by progressively refining extracted instances through clustering and confidence computation at each hop, rather than attempting extraction in a single step.
Solution Approach 2:
The patent employs preliminary action by using manually curated seed instances (both positive and negative seeds) before the main extraction process. These seeds serve as reliable starting points that guide the iterative extraction and clustering process, improving the reliability of subsequently extracted instances by anchoring them to verified examples.
2Measurement precision
If iterative bootstrapping with multiple hops is employed, then the hit rate and reliability improve, but the processing time and computational resources increase
Solution Approach 1:
The patent applies partial action by allowing the iterative bootstrapping process to be stopped after a predetermined number of hops or when convergence is achieved. This means the system performs only as many extraction and clustering iterations as necessary to achieve satisfactory results, rather than always completing a fixed number of hops, thus reducing processing time while maintaining accuracy.
Solution Approach 2:
The patent implements feedback mechanisms through confidence computation at each hop, where the reliability of extracted instances is evaluated and used to guide subsequent extraction steps. This feedback allows the system to adjust its processing dynamically, stopping or continuing based on the quality of extracted instances, thereby optimizing the balance between accuracy and processing time.
3Reliability
If clustering and confidence computation are performed at each hop, then false positives are reduced, but the computational complexity increases
Solution Approach 1:
The patent segments the complex processing into distinct modular steps: extraction based on seeds, clustering of extracted instances, confidence computation for each cluster, and selection of high-confidence instances for the next hop. This segmentation makes the overall complex process more manageable and allows each component to be optimized independently while maintaining high reliability.
Solution Approach 2:
The patent changes parameters dynamically across hops, including the number of seeds, clustering thresholds, and confidence levels. By adjusting these parameters based on the results of previous hops, the system adapts to the data characteristics and maintains high reliability without requiring excessively complex fixed processing rules.
Data Source
AI summary
Present invention concerns a method of relation extraction from a text corpus, the method comprising extracting instances from the text corpus based on seeds, wherein the seeds include at least one set of template seeds and at least one set of entity seeds. The invention also pertains to related devices and methods.