Knowledge Graph Question Answering for Accurate Multi-Hop Queries
Find Innovative SolutionsGenerate Solutions
Solution Overview
Problem
Existing question answering models, such as LLMs and supervised learning methods, generate inaccurate and inefficient answers due to reliance on unstructured data, hallucinations, and the high cost of labeled data, leading to inefficient resource usage and incorrect information.
Innovation Solution
A computer-implemented method that generates a query focused hyper-relational knowledge graph by extracting entities and relations from input documents, using a pretrained neural network to predict answers without labeled data, optimizing storage and computational resources.
Engineering Contradictions & Design Principles
Engineering Contradiction Analysis
1Ease of operation
If large language models are used for question answering, then structured sentence generation capability is improved, but factual accuracy deteriorates due to hallucinations
Solution Approach 1:
The patent introduces an intermediary verification mechanism between the LLM's sentence generation and the final output. A fact-checking module acts as a mediator that validates the generated sentences against the knowledge base, preventing hallucinated facts from being output while preserving the LLM's natural language generation capabilities.
Solution Approach 2:
The system implements a feedback loop where the output of the LLM is continuously validated against the knowledge base. When inconsistencies or hallucinations are detected, the system provides feedback to correct the generated sentences, ensuring factual accuracy while maintaining the ease of operation of the LLM.
2Reliability
If training-based methods are used for multi-hop question answering, then answer accuracy is improved, but data labeling cost increases
Solution Approach 1:
The system enables self-service through automated entity extraction and relation identification from unstructured text. The knowledge graph is automatically constructed from available documents without requiring manual labeling, allowing the system to achieve high answer accuracy while eliminating the costly data labeling process.
Solution Approach 2:
The patent performs preliminary action by pre-processing documents to extract entities and relations before the actual question answering process. This preliminary structuring of information into a knowledge base eliminates the need for time-consuming data labeling during training, while maintaining high accuracy for multi-hop questions.
3Loss of information
If complete text documents are accessed for question answering, then information completeness is improved, but computational efficiency deteriorates
Solution Approach 1:
The system extracts only the relevant entities and relations from complete text documents to construct a knowledge graph. This extraction process removes unnecessary information while preserving all factually relevant data, enabling the system to maintain information completeness while significantly improving computational efficiency during question answering.
Solution Approach 2:
The patent segments the complete text documents into discrete entities and relations, organizing them into a structured knowledge graph. This segmentation allows the system to access only the specific segments relevant to each question, rather than processing entire documents, thereby maintaining information completeness while enhancing computational efficiency.
Data Source
Figure 1
Figure 2
Figure 3
AI summary
A computer-implemented method of generating an answer from an input query and input documents, comprising extracting input query entities from the input query and input document entities from the input documents, sampling a schema of in-domain queries with the input query to generate a query sampled schema, generating an entity-document graph from the input documents and input document entities, generating a hyper-relational knowledge graph by extracting, for each input query entity, a document title and relation to an input document entity of the input document entities from the input documents in the entity-document graph, sampling the hyper-relational knowledge graph with the query sampled schema to generate a query focused hyper-relational knowledge graph, predicting an answer to the input query by inputting the query focused hyper-relational knowledge graph and input query into a pretrained neural network, outputting the answer.