Question Answering via Feature Vector Encoding and Sub-Graph Retrieval
Find Innovative SolutionsGenerate Solutions
Solution Overview
Problem
Current Question Answering Systems (QA) face challenges in providing high-quality answers due to reliance on manual construction of standard questions and answers, limiting flexibility and accuracy in responding to user queries.
Innovation Solution
A method involving encoding user questions and pre-constructed structured data to obtain feature vectors, retrieving candidate sub-graphs, and performing multi-task classification to generate answer content, which allows for flexible and personalized question answering without fixed mappings between questions and answers.
Engineering Contradictions & Design Principles
Engineering Contradiction Analysis
1Reliability
If manual construction of standard questions and answers is used, then answer quality can be maintained through human expertise, but system flexibility and adaptability to new queries deteriorate
Solution Approach 1:
The system performs self-service by automatically encoding user questions into feature vectors and retrieving relevant sub-graphs from structured knowledge data without requiring manual construction of standard questions. The multi-task classification model enables the system to adapt to diverse query types autonomously
Solution Approach 2:
The patent transforms the parameter representation by encoding questions and structured data into feature vectors, changing the state from discrete manual categories to continuous vector space representations. This enables flexible adaptation while maintaining answer quality through the learned parameter transformations
2Stability of the object's composition
If fixed mappings between questions and answers are established, then answer consistency is improved, but system complexity and difficulty in handling personalized queries increase
Solution Approach 1:
The patent replaces the mechanical system of fixed question-answer mappings with a neural network-based feature vector encoding system. The multi-task classification model substitutes rigid lookup tables with learned representations that maintain consistency through shared encoding while handling diversity through task-specific classification heads
3Reliability
If comprehensive structured knowledge data is processed, then answer accuracy and knowledge richness are improved, but processing time and computational resources increase
Solution Approach 1:
The system extracts only the relevant sub-graphs from the comprehensive structured knowledge data based on the encoded question features. This extraction principle filters out unnecessary information, maintaining answer accuracy through targeted retrieval while reducing processing time by avoiding full-data scanning
Solution Approach 2:
The patent performs preliminary encoding of both user questions and structured knowledge data into feature vectors before retrieval. This preliminary action prepares the data in an optimized format that enables faster matching and reduces computational burden during the actual query processing phase
Data Source
AI summary
Embodiments of the present specification disclose a method, an apparatus, a system, a device, and a storage medium for answering user questions, including: obtaining a user question; encoding the user question and a schema level of pre-constructed structured data to obtain a first feature vector, wherein the structured data further comprises a data level, wherein the data level comprises knowledge for answering questions structured according to the schema level; retrieving one or more candidate sub-graphs related to the user question from the structured data; encoding the one or more candidate sub-graphs to obtain a second feature vector; performing multi-task classification for the user question based on the first feature vector and the second feature vector; and obtaining answer content for the user question based on a result of the multi-task classification.