Stance Classification for Multi-Perspective Consumer Health Queries
Find Innovative SolutionsGenerate Solutions
Solution Overview
Problem
Current search engines are ineffective in addressing complex consumer health queries with no single definitive answer, as they fail to classify the diverse perspectives supporting or opposing the query, making it challenging for users to synthesize a balanced view, especially in the medical domain where language is technical and lacks emotional cues.
Innovation Solution
A method and system for automatic stance classification using SVM-SC and NN-SC approaches, which extract stance vectors, biomedical semantic relations, and textual entailment features to classify propositions as supporting, opposing, or neutral to a query, providing a balanced view for decision-making.
Engineering Contradictions & Design Principles
Engineering Contradiction Analysis
1Loss of information
If general search engines are used to address complex consumer health queries, then information retrieval is performed, but the ability to classify diverse perspectives and provide balanced views deteriorates
Solution Approach 1:
The patent segments health information into distinct proposition units with defined stances (support, oppose, neutral). Each proposition is independently classified and tagged with its stance label, enabling users to systematically view different perspectives on health queries rather than encountering undifferentiated search results.
Solution Approach 2:
The patent introduces an intermediary classification system that acts as a mediator between raw health information and user comprehension. The stance classification framework serves as this intermediary, organizing diverse perspectives into structured categories that facilitate balanced view synthesis while preserving all original information.
2Measurement precision
If stance classification framework with multiple features is implemented, then classification accuracy improves, but system complexity increases
Solution Approach 1:
The classification system is segmented into distinct modular components: stance vector extraction module, biomedical semantic relations extraction module, textual entailment extraction module, and classification module. Each component performs a specific function and can be independently developed and optimized, reducing overall system complexity while maintaining high accuracy.
Solution Approach 2:
The patent creates a universal stance classification framework that can handle multiple types of health information and queries through a single integrated system. The framework uses multiple feature extraction methods (stance vectors, semantic relations, textual entailment) that work together to provide comprehensive classification across diverse medical domains.
Data Source
AI summary
Method, system, and apparatus for automatic stance classification. Propositions can be collected that are relevant to a query. A classifier can classify the stance of each proposition based on whether the proposition supports the query, opposes the query, or is neutral with respect to the query in order to thereafter provide substantive data for decision making based on and extracted from the query. The stance can be classified based on, for example, an SVM-SC (SVM Based Stance Classification) approach and/or an NN-SC (Neural Network Stance Classification Approach).