Synonymous Expression Assessment Using Concept-Specific Distributions
Find Innovative SolutionsGenerate Solutions
Solution Overview
Problem
Existing methods for assessing synonymous expressions of binary relations face challenges when input predicates or nominals are polysemous, as they struggle to acquire sufficient feature values and assess similarity accurately.
Innovation Solution
A device and method that compute similarity between input predicates and nominals using distributions of occurrence frequencies of nominals and predicates within the same concept type, ensuring accurate assessment even with polysemous inputs.
Engineering Contradictions & Design Principles
Engineering Contradiction Analysis
1Measurement precision
If existing methods collect surrounding contexts of binary relations as feature values, then synonymous expressions can be extracted, but insufficient feature values are acquired when sentences contain only binary relations
Solution Approach 1:
The patent introduces concept types as an intermediary layer between predicates and nominals. By computing concept types for each predicate and nominal, and using distributions of nominals within the same concept type as feature values, the system can extract synonymous expressions even from sentences containing only binary relations, without relying on surrounding contexts.
2Measurement precision
If existing methods use distribution of occurrence frequencies of all nominals for predicate similarity, then comprehensive feature values are obtained, but accuracy decreases when predicates or nominals are polysemous
Solution Approach 1:
The patent segments the set of all nominals into subsets based on concept types. Instead of computing similarity using distributions of all nominals, the system computes similarity using distributions of nominals that belong to the same concept type as the input nominal. This segmentation eliminates interference from polysemous nominals of different meanings while maintaining computational feasibility.
3Measurement precision
If existing methods assess predicate and nominal similarity separately, then comprehensive assessment is achieved, but polysemous inputs produce inaccurate results due to mixed concept distributions
Solution Approach 1:
The patent applies local quality by making the feature value computation adaptive to the specific concept type of each input nominal. The system computes distributions of nominals locally within the same concept type, ensuring that the feature values reflect only conceptually relevant information. This local computation preserves concept-specific information that would be lost in global distributions.
Data Source
AI summary
A synonymous expression assessment device includes: synonymy assessment means for receiving input of binary relations each of which includes a nominal and a predicate, and assessing whether or not the input binary relations are synonymous using a similarity between input nominals and a similarity between input predicates; and inter-predicate similarity computation means for, when computing the similarity between the input predicates based on a distribution of occurrence frequencies of nominals that are in binary relations to the input predicate in a document set, performing the computation using a distribution of only nominals that are used in the same type of concept as the input nominal.