Question-Answering Confidence Model Using Iterative Feature Extraction
Find Innovative SolutionsGenerate Solutions
Solution Overview
Problem
Existing question-answering systems fail to provide accurate answers when the correct answer is not included in the candidate answer set, as the data is not related to the machine learning model, leading to poor resolution.
Innovation Solution
A machine learning method that receives questions, forms training data sets using feature sets from candidate answers, performs machine learning to compute confidence degrees, and iteratively refines these sets to improve answer accuracy, using both initial and secondary feature sets to enhance answer determination.
Engineering Contradictions & Design Principles
Engineering Contradiction Analysis
1Reliability
If traditional machine learning methods are used with candidate answer sets, then the system can process questions with available answers, but it fails to provide accurate answers when the correct answer is not included in the candidate answer set
Solution Approach 1:
The patent performs preliminary actions by pre-training the machine learning model with extensive training data that includes various answer patterns and contexts. This preliminary training enables the model to recognize and evaluate candidate answers more effectively, even when the correct answer was not explicitly present in the training set, thereby improving reliability for unseen questions.
Solution Approach 2:
The patent changes key parameters of the machine learning model, specifically adjusting the confidence threshold and incorporating multiple feature sets (including answer relevance, question-answer matching, and contextual features). By dynamically adjusting these parameters based on the confidence score, the system can adapt to different question types and improve its ability to identify correct answers even when they are not initially obvious in the candidate set.
2Measurement precision
If the system uses a single feature set for machine learning, then the model is simpler to train, but the resolution and accuracy of answer identification is insufficient
Solution Approach 1:
The patent segments the feature extraction process into multiple distinct feature sets, each capturing different aspects of the question-answer relationship. The first feature set focuses on basic matching, the second on contextual relevance, and the third on semantic similarity. This segmentation allows the model to process each feature type separately and combine them, improving measurement precision while managing complexity through modular processing.
Solution Approach 2:
The patent adds another dimension to the feature space by incorporating multiple feature sets with different granularities and perspectives. Instead of relying on a single flat feature vector, the system creates a multi-dimensional feature space that includes syntactic, semantic, and contextual dimensions. This dimensional expansion enables more precise answer identification by capturing nuanced relationships that a single feature set would miss.
Data Source
AI summary
A method and system for computer question-answering. The method includes: receiving a plurality of the questions to be trained, and obtaining a candidate answer set for each question; determining part of the questions to which the candidate answer sets include correct answers, and using first feature sets of the candidate; obtaining a first mathematical model, which is used to compute a first confidence degree that the candidate answer is a correct answer; computing first confidence degrees of the candidate answers of all the questions to be trained, and for each question, extracting a second feature set related to the first confidence degrees and the first feature set; and obtaining a second mathematic model, which is used to compute a second confidence degree that the candidate answer is a correct answer based on the second feature set.