Dependency Syntax Parsing for Event Information Extraction
Find Innovative SolutionsGenerate Solutions
Solution Overview
Problem
Current Named Entity Recognition (NER) and Entity Linking (EL) technologies can recognize and link entities in text but are unable to effectively recognize and link events.
Innovation Solution
A method and apparatus that performs dependency syntax parsing on a target text to generate a dependency tree, matches syntactic structure trees with the dependency tree to obtain triples, and determines a target triple based on word content and predetermined weights, allowing for the identification of relevant event information.
Engineering Contradictions & Design Principles
Engineering Contradiction Analysis
1Measurement precision
If NER and EL technologies are used to mine text for entities, then entity recognition and linking capability is improved, but event recognition and linking capability remains impossible
Solution Approach 1:
The patent segments the text processing task into multiple stages: first performing dependency syntax parsing to generate a dependency tree, then matching syntactic structure trees to extract candidate triples, and finally selecting target triples through scoring. This segmentation allows the system to handle both entity and event recognition systematically.
Solution Approach 2:
The patent introduces dependency trees and syntactic structure trees as intermediary representations between raw text and event information. These intermediate structures serve as mediators that transform unstructured text into organized triplets (subject, predicate, object), enabling event recognition while building upon existing NER capabilities.
2Loss of information
If multiple syntactic structure trees are matched with dependency tree to obtain multiple triples, then event information coverage is improved, but target triple selection difficulty increases
Solution Approach 1:
The patent changes the parameters of triple evaluation by introducing a scoring mechanism that combines multiple factors: syntactic structure tree matching weights, word co-occurrence frequencies, and positional relationships. This parameter-based scoring system transforms the complex selection problem into a quantitative comparison task.
Solution Approach 2:
The patent replaces manual or rule-based triple selection with an automated scoring system that calculates composite scores based on predefined criteria. This substitution of mechanical selection processes with algorithmic evaluation reduces complexity while maintaining comprehensive event information coverage.
3Measurement precision
If dependency syntax parsing and syntactic structure matching are performed, then event extraction accuracy is improved, but processing time and computational complexity increase
Solution Approach 1:
The patent performs dependency syntax parsing and generates dependency trees as a preliminary step before event extraction. By preparing this structured representation in advance, the system enables more efficient event pattern matching and triple extraction, reducing the computational burden during the actual event recognition phase.
Solution Approach 2:
The dependency tree structure serves multiple functions: it captures grammatical relationships for accurate event extraction, provides a framework for matching syntactic patterns, and enables subsequent information retrieval tasks. This multi-functionality reduces the need for separate processing steps.
Data Source
AI summary
A method and apparatus for generating information are disclosed. An implementation of the method includes: receiving a target text, the target text including an objective and descriptive information of the objective; performing a dependency syntax parsing on the target text to generate a dependency tree of the target text; matching predetermined syntactic structure tree with the dependency tree to obtain at least one triple, a triple including a subject, a predicate, and an object; and determining, based on words contained in a triple among the at least one triple and a predetermined weight of the syntactic structure tree matched to obtain the triple, a target triple among the at least one triple.