Cognitive System Question Variant Generation
Find Innovative SolutionsGenerate Solutions
Solution Overview
Problem
The time-consuming process of generating and recognizing equivalent questions for cognitive systems, such as customer help desk tools, hinders the efficient training of these systems, as they require numerous variations of the same question to optimize performance effectively.
Innovation Solution
A computer-implemented method that applies rules of grammar and common usage to generate variants of training questions, identifies generalized words, and uses a classifier to filter out nonsensical variants, providing synonymous questions that are equivalent or different forms of the same question.
Engineering Contradictions & Design Principles
Engineering Contradiction Analysis
1Reliability
If questions are manually generated for training cognitive systems, then the system can learn from experience, but the training process becomes extremely time-consuming
Solution Approach 1:
The patent applies copying by generating multiple variant forms of training questions from a single source question. The question generation component creates syntactically and semantically equivalent questions (e.g., passive/active voice transformations, synonym replacements) to expand the training dataset automatically, eliminating the need for manual creation of numerous question variants while maintaining training effectiveness
Solution Approach 2:
The patent implements preliminary action by pre-generating comprehensive sets of question variants before the actual training process begins. The system creates all possible syntactic and semantic variations of training questions in advance, so that when training starts, the cognitive system immediately has access to a diverse question set without requiring time-consuming manual generation during the training phase
2Adaptability or versatility
If numerous question variants are generated to improve system performance, then the system recognizes more equivalent questions, but the complexity of generating and filtering variants increases
Solution Approach 1:
The patent applies segmentation by dividing the question generation process into distinct modular components: a question generation component that creates variants using syntactic transformation rules, a classifier component that filters nonsensical variants, and a training component that uses the filtered questions. This modular architecture manages complexity by allowing each component to specialize in one function while working together to achieve comprehensive question variant generation
Solution Approach 2:
The patent implements feedback through the classifier component that evaluates generated question variants and provides feedback by filtering out nonsensical ones. The classifier uses machine learning to assess whether generated variants are meaningful, creating a feedback loop that ensures only high-quality question variants are used for training, thereby improving question recognition capability without requiring manual review of every variant
Data Source
AI summary
A computer-implemented method, apparatus and computer program product of question generation in a cognitive system, comprising: receiving a training question; applying rules of grammar and common usage to generate variants of the training question; identifying any generalized versions of specific words, such as verbs or nouns in the training question; identifying any questions related to the training question; filtering out, utilizing a classifier, generated variants which do not make sense; and providing one or more output questions, synonymous to the training question.