Tailored Knowledge Graph Induction for Interpretable Question Answering
Find Innovative SolutionsGenerate Solutions
Solution Overview
Problem
Existing question answering techniques using knowledge graphs are either opaque and poorly equipped for complex reasoning or lack transparency and efficiency, while deep learning methods are not easily interpretable and struggle with complex queries.
Innovation Solution
A hybrid approach that iteratively constructs a tailored knowledge graph from structured sub-queries, leveraging deep learning to fill parameter slots and generate a tailored knowledge graph relevant to the question, enhancing both interpretability and complex reasoning capabilities.
Engineering Contradictions & Design Principles
Engineering Contradiction Analysis
1Ease of operation
If existing knowledge graph techniques are used for question answering, then interpretability is improved, but complex reasoning capability deteriorates
Solution Approach 1:
The patent segments the question answering process into multiple structured sub-queries that are processed iteratively. Each sub-query focuses on extracting specific information from the corpus and building particular aspects of the knowledge graph, allowing complex reasoning to be broken down into manageable, interpretable steps while maintaining overall reasoning capability.
Solution Approach 2:
The knowledge graph is constructed dynamically and iteratively based on the specific question being asked, rather than using a static pre-curated graph. The system adapts the knowledge graph structure and content to the specific reasoning needs of each query, enabling complex reasoning while maintaining interpretability through question-specific tailoring.
2Reliability
If pre-curated knowledge graphs are used, then complex reasoning is improved, but information relevance deteriorates
Solution Approach 1:
The patent extracts only the specific information and entities relevant to the given question from the corpus to construct the knowledge graph. Rather than using all information from a pre-curated graph, the system selectively extracts and includes only what is necessary for answering the specific question, eliminating irrelevant information while preserving reasoning capability.
Solution Approach 2:
The system performs preliminary analysis of the question to identify required information types and structures before constructing the knowledge graph. This preliminary action allows the system to build a tailored knowledge graph with only the necessary components for complex reasoning, avoiding inclusion of irrelevant information from the outset.
3Productivity
If deep learning pipelines are used for question answering, then performance is improved, but interpretability deteriorates
Solution Approach 1:
The patent introduces a knowledge graph as an intermediary between the deep learning pipeline and the final answer. The deep learning model generates structured sub-queries and extracts information, which is then organized into an interpretable knowledge graph structure. This intermediary representation maintains the high performance of deep learning while providing transparent, structured output that can be easily interpreted and verified.
4Reliability
If all corpus information is processed, then completeness is improved, but processing efficiency deteriorates
Solution Approach 1:
The system extracts only the specific portions of the corpus that are relevant to answering the given question, rather than processing the entire corpus. By identifying and extracting only necessary information through structured sub-queries, the system maintains completeness of the answer while dramatically improving processing efficiency by avoiding unnecessary computation on irrelevant data.
Data Source
AI summary
Systems and techniques that facilitate question answering via tailored knowledge induction are provided. In various embodiments, a system can access a natural language question. In various aspects, the system can identify an answer for the natural language question, based on a tailored knowledge graph iteratively generated from a set of structured sub-queries associated with the natural language question.