Machine Learning Framework for Double-Barreled Question Detection
Find Innovative SolutionsGenerate Solutions
Solution Overview
Problem
Double-barreled questions in surveys, which address multiple subjects with a single answer option, lead to response bias due to unclear answer application, and existing methodologies struggle to accurately classify and correct such questions, often misflagging properly constructed questions.
Innovation Solution
A machine learning framework is implemented in an online survey platform to classify and interpret double-barreled questions, addressing data imbalance through active learning and optimal model selection, and providing token importance visualization to explain predictions.
Engineering Contradictions & Design Principles
Engineering Contradiction Analysis
1Ease of manufacture
If a simple grammatical conjunction approach is used to identify double-barreled questions, then the identification process is straightforward and easy to implement, but the accuracy deteriorates because properly constructed questions may have conjunctions leading to incorrect flagging
Solution Approach 1:
The patent replaces the mechanical/grammatical approach of detecting conjunctions with a machine learning-based semantic analysis system. The system uses trained models to analyze the meaning and structure of survey questions, substituting simple rule-based mechanics with intelligent computational analysis that can distinguish between legitimate conjunctions and actual double-barreled questions.
Solution Approach 2:
The patent changes the parameters of question analysis from surface-level grammatical features to deeper semantic and contextual features. By training machine learning models on labeled data, the system learns to evaluate multiple parameters including question structure, answer options, and semantic relationships, transforming the identification process from simple conjunction detection to comprehensive semantic analysis.
2Productivity
If machine learning models are trained with imbalanced data sets, then the training process is simpler and faster, but the model performance deteriorates due to bias towards the majority class
Solution Approach 1:
The patent intentionally introduces asymmetry into the training process by applying different sampling strategies to minority and majority classes. The system uses oversampling techniques for minority class samples and undersampling for majority class samples, creating an asymmetric training approach that balances the effective representation of both classes and improves model performance on imbalanced data.
Solution Approach 2:
The patent implements dynamic resampling strategies that adapt during the training process. The system dynamically adjusts the sampling rates and class weights based on the current training state and model performance, allowing the training process to evolve from simple imbalanced data to a more balanced effective representation as training progresses.
3Measurement precision
If comprehensive feature extraction is performed to improve double-barreled question detection accuracy, then the identification precision improves, but the computational complexity and processing time increase
Solution Approach 1:
The patent performs preliminary feature extraction and text processing during the model training phase rather than during real-time inference. By pre-processing and extracting comprehensive features during training when computational resources are available, the system reduces the computational burden during actual survey question analysis, achieving high accuracy without excessive processing time.
Solution Approach 2:
The patent segments the feature extraction process into multiple stages: tokenization, part-of-speech tagging, dependency parsing, and semantic role labeling. This segmentation allows the system to apply computational complexity selectively at different processing stages, optimizing the balance between detection accuracy and processing efficiency by only performing full feature extraction where needed.
Data Source
AI summary
A computer-implemented method includes gathering data samples into a data set, correcting for imbalance in the data set to produce a corrected data set by applying active learning to the data set to increase a number of double barreled question data samples occurring in the data set, selecting an optimal machine learning model for the corrected data set, training the optimal machine learning model using the corrected data set, operating the optimal machine learning model on new data to produce a prediction result, and generating a visual representation of at least one prediction results.