Causal Learning Data Generation With Validated Negative Examples
Find Innovative SolutionsGenerate Solutions
Solution Overview
Problem
Existing methods for generating negative examples of causal relationship expressions in text data suffer from the acquisition of inappropriate examples due to insufficient examination, leading to potential inaccuracies.
Innovation Solution
A learning data generation device and method that involves extracting cause and result expressions from input text, modifying them through interchange or replacement with dissimilar sentences, and validating the appropriateness of the modified examples using external databases to ensure accuracy.
Engineering Contradictions & Design Principles
Engineering Contradiction Analysis
1Ease of manufacture
If negative examples are generated by randomly replacing cause or result expressions in positive examples, then the data collection process is simple and convenient, but the generated negative examples may be inappropriate or not true negatives
Solution Approach 1:
The patent applies preliminary action by extracting cause and result expressions from positive examples before generating negative examples. The system pre-processes the positive examples to identify the causal structure, then uses this extracted structure to generate appropriate negative examples by replacing expressions while maintaining grammatical correctness and contextual appropriateness, thereby avoiding the generation of inappropriate negatives.
Solution Approach 2:
The patent implements feedback by using a judgment database that stores appropriate negative examples for reference. The system compares generated negative examples against this database to verify their appropriateness, and can use this feedback to adjust the generation process, ensuring that only appropriate negative examples are produced while maintaining the simplicity of the overall process.
2Productivity
If simple replacement methods are used to generate negative examples, then the generation process is efficient, but data quality deteriorates due to insufficient examination
Solution Approach 1:
The patent applies segmentation by dividing the negative example generation process into distinct stages: (1) extracting cause and result expressions from positive examples, (2) generating candidate negative examples through controlled replacement, and (3) judging the appropriateness of generated examples against the judgment database. This segmentation allows the system to maintain high productivity while improving quality through structured processing and verification.
Solution Approach 2:
The patent uses an intermediary approach by introducing a judgment database as a mediator between the generation process and the final output. This database serves as a reference standard that enables quality control without significantly slowing down the generation process, allowing the system to efficiently produce high-quality negative examples by comparing candidates against pre-stored appropriate examples.
Data Source
AI summary
A learning data generation device includes processing circuitry to extract a cause expression and a result expression from an input text, and to generate a modified text by at least one of a method of interchanging the cause expression and the result expression and a method of specifying one of the cause expression and the result expression as a modification target sentence and replacing the modification target sentence with a replacement candidate sentence dissimilar to the modification target sentence.