Natural Language Question Answering via Linked Data URI Spreading
Find Innovative SolutionsGenerate Solutions
Solution Overview
Problem
Current Question Answering (QA) systems are limited in processing natural language questions over open linked structured information, as they rely on specific natural language grammar and fail to effectively utilize the non-ambiguous characteristics of structured data, particularly in the context of open linked data.
Innovation Solution
A method and apparatus that detect named entities in natural language questions, extract relevant information, search linked data using Uniform Resource Identifiers (URIs) and spreading activation, generate candidate answers, and evaluate them by synthesizing feature values, allowing for the processing of natural language questions with open domain and free grammar.
Engineering Contradictions & Design Principles
Engineering Contradiction Analysis
1Reliability
If NLDB techniques translate natural language questions into executable SQLs based on database schema syntax, then the system can process structured data, but users are restricted to ask questions with specific natural language grammar
Solution Approach 1:
The patent introduces an intermediary layer (question analysis module and mapping module) between the natural language input and the database query execution. This intermediary translates free-grammar natural language questions into structured SQL queries by analyzing semantic relationships and mapping them to database schemas, thereby resolving the contradiction between maintaining grammatical flexibility and ensuring query accuracy
Solution Approach 2:
The system segments the question processing into distinct modules: question analysis for extracting semantic information, mapping for translating to SQL syntax, and execution for running queries. This segmentation allows each module to specialize in handling either natural language variability or structured data requirements, resolving the contradiction between grammatical flexibility and answer accuracy
2Adaptability or versatility
If QA systems use unstructured data with NLP techniques, then they can process over 80% of world data, but they fail to effectively utilize the non-ambiguous characteristics of structured data
Solution Approach 1:
The system dynamically adapts its processing approach based on the data type. For structured data, it employs precise SQL-based processing that leverages the non-ambiguous characteristics of structured information. For unstructured data, it uses NLP techniques. This dynamic adaptation resolves the contradiction by selecting the appropriate processing method based on data characteristics
Solution Approach 2:
The patent introduces an intermediary layer that bridges unstructured natural language questions and structured database data. This intermediary analyzes the semantic meaning of questions and maps them to structured queries, enabling the system to effectively utilize structured data while maintaining the ability to process natural language input
3Measurement precision
If the system searches in multiple linked data sources and merges candidate answers, then the precision of QA systems is improved, but the complexity of the processing increases
Solution Approach 1:
The system implements feedback mechanisms where candidate answers from multiple data sources are evaluated and ranked based on their relevance and accuracy. The evaluation module uses feedback from the matching process to refine the selection, ensuring that the most accurate answer is chosen while managing processing complexity through systematic evaluation criteria
Data Source
AI summary
A method and an apparatus for selecting an answer to a natural language question. The method includes: detecting a named entity in the natural language question; extracting information related to an answer from the natural language question; searching in linked data according to the detected named entity; generating a candidate answer according to a search result; parsing the candidate answer according to the information related to the answer; and obtaining a value of a feature of the candidate answer; and evaluating each candidate answer by synthesizing the value of the feature of the candidate answer.