Geographic Relevance Scoring in Deep Question Answering
Find Innovative SolutionsGenerate Solutions
Solution Overview
Problem
Existing question answering systems face challenges in accurately identifying and ranking answers that contain geographical information, as they often fail to recognize and utilize the geographic focus of questions, leading to incorrect answers when questions include geographical details.
Innovation Solution
A system and method that analyze input natural language questions to identify their geographic focus, process documents to generate candidate answers, and evaluate the geographic relevance of these answers by comparing their focus with the question's focus, using document location data and normalization of locative expressions to improve answer accuracy.
Engineering Contradictions & Design Principles
Engineering Contradiction Analysis
1Device complexity
If geographic information is ignored in question answering systems, then the system complexity is reduced, but the answer accuracy deteriorates when questions contain geographical details
Solution Approach 1:
The patent extracts geographic information from questions and answer candidates as a separate feature dimension. By identifying and isolating geographic entities and their relationships, the system can evaluate geographic relevance independently without complicating the overall QA architecture, thus maintaining answer accuracy while avoiding excessive system complexity
Solution Approach 2:
The patent introduces geographic focus as an intermediary concept that mediates between the question and answer candidate. This intermediary allows the system to evaluate geographic relevance without directly complicating the core QA matching process, serving as a bridge that improves accuracy while keeping the system architecture manageable
2Manufacturing precision
If geographic focus identification is implemented, then the answer accuracy for geographical questions is improved, but the processing complexity increases
Solution Approach 1:
The patent performs preliminary identification of geographic focus in both questions and answer candidates before the main matching process. By pre-processing and tagging geographic entities and their relationships in advance, the system avoids complex real-time analysis during question answering, thus improving accuracy while managing processing complexity
Solution Approach 2:
The patent segments the question and answer candidate processing into distinct components: geographic entity identification, geographic focus determination, and geographic relevance evaluation. This segmentation allows each component to be optimized independently, improving overall accuracy while keeping individual processing steps manageable
3Measurement precision
If geographic relevance evaluation is performed for all candidate answers, then the precision of geographical answer matching is improved, but the computational time increases
Solution Approach 1:
The patent applies geographic relevance evaluation selectively based on local characteristics of questions and answer candidates. By identifying questions that contain geographic information and focusing evaluation only on relevant candidates, the system achieves high geographic matching precision without performing unnecessary evaluations on all answers, thus reducing computational time
Data Source
AI summary
A method, system and a computer program product are provided for scoring candidate answers for geographic relevance by identifying document location information that is associated with a document, associating each token in the document with the document location information, and then comparing geographic foci identified for a candidate answer from the tokens with geographic foci identified for an input question to generate a geographic relevance score for the candidate answer to the input question.