Automatic Question Answering Corpus Expansion via Candidate Evaluation
Find Innovative SolutionsGenerate Solutions
Solution Overview
Problem
Domain-specific Question Answering (QA) systems face challenges in accurately handling queries due to limited corpora, leading to incorrect or incomplete answers, especially when queries relate to background information not covered by the corpus, resulting in potential revenue loss and increased system complexity.
Innovation Solution
A computer-implemented method to automatically expand the QA system's corpus by identifying portions for expansion, generating search queries based on semantic and linguistic features, and integrating candidate answers that meet predetermined quality ratings, leveraging semantic resources and graph-based techniques to ensure relevance and accuracy.
Engineering Contradictions & Design Principles
Engineering Contradiction Analysis
1Reliability
If the corpus is manually expanded with hand-written data, then the quality of corpus content is improved, but the cost and time investment increase significantly
Solution Approach 1:
The QA system automatically evaluates and selects its own corpus expansion needs by generating search queries, retrieving candidate answers, and assessing their quality against existing content, eliminating the need for manual expert evaluation while maintaining high content quality
Solution Approach 2:
The system changes the parameter of corpus expansion from manual expert-driven process to automated algorithm-driven process, using quality thresholds and similarity metrics to automatically determine which candidate content should be added to the corpus
2Quantity of substance
If the corpus is automatically expanded with external data, then the quantity of corpus content increases, but the relevance and quality of added content decrease
Solution Approach 1:
The system uses feedback loops where candidate answers are evaluated against existing corpus content using similarity metrics and quality thresholds, automatically rejecting irrelevant or low-quality content while accepting only those that meet predetermined standards
Solution Approach 2:
The patent replaces manual expert judgment with automated computational methods including similarity algorithms, quality scoring mechanisms, and threshold-based filtering to objectively assess candidate content relevance and quality
3Adaptability or versatility
If too much data is added to the corpus, then the coverage of queries improves, but the system complexity and computational resources increase
Solution Approach 1:
The system applies local quality control by evaluating each candidate answer individually against specific criteria (similarity to existing content, quality thresholds, relevance to domain), adding only the necessary portions of content rather than bulk importing data
4Measurement precision
If the corpus is expanded strategically with high-quality related data, then the answer accuracy improves, but the manual process becomes time-consuming and expensive
Solution Approach 1:
The QA system performs self-evaluation by automatically generating search queries, retrieving candidate answers, assessing their quality and relevance, and making decisions about corpus expansion without human intervention, thereby maintaining high answer accuracy while dramatically improving productivity
Data Source
AI summary
Expanding the corpus of a question answering system. The question answering system is adapted to generate a candidate answer to a query. The candidate answer may be incorporated into the corpus of the question answering system if the candidate answer meets or exceeds a predetermined requirement. By expanding the corpus with material that has been determined to be useful for answering a query, for example, the corpus may be expanded automatically in an accurate and efficient manner.