Question Answering with Query-Focused Hyper-Relational Knowledge Graphs
Find Innovative SolutionsGenerate Solutions
Solution Overview
Problem
Existing question answering models, such as large language models (LLMs) and supervised learning methods, face challenges in generating accurate and contextually meaningful answers due to the need for large amounts of labeled data, inefficiencies in resource usage, and the generation of conflicting information from unstructured text.
Innovation Solution
A computer-implemented method involving the extraction of input query and document entities, generation of a query sampled schema, entity-document graph, and hyper-relational knowledge graph, followed by sampling and prediction using a pretrained neural network to generate answers without requiring labeled data.
Engineering Contradictions & Design Principles
Engineering Contradiction Analysis
1Reliability
If training-based methods are used for multi-hop question answering, then answer accuracy is improved, but large amounts of labeled data are required during training
Solution Approach 1:
The patent extracts and separates the knowledge representation from the training data requirement by using knowledge graphs to store structured information. This allows the system to leverage existing structured knowledge without requiring extensive labeled training data for each specific question answering task.
Solution Approach 2:
The patent introduces knowledge graphs as an intermediary between the question answering system and the data source. The knowledge graph serves as a mediator that structures information in a way that can be queried directly, eliminating the need for traditional training-based approaches that require large labeled datasets.
2Ease of operation
If large language models are used for question answering, then structured sentence generation is improved, but hallucination of facts occurs
Solution Approach 1:
The patent introduces knowledge graphs as an intermediary between the question answering system and the data source. The knowledge graph serves as a mediator that structures information in a way that can be queried directly, eliminating the need for traditional training-based approaches that require large labeled datasets.
Solution Approach 2:
The patent replaces the probabilistic generation mechanism of LLMs with a deterministic query-based retrieval mechanism using knowledge graphs. This substitution eliminates hallucination by ensuring answers are derived from verified structured knowledge rather than probabilistic predictions.
3Loss of information
If whole text documents are accessed for question answering, then complete information is obtained, but computing resources are inefficiently used
Solution Approach 1:
The patent segments the information retrieval process by dividing the knowledge base into structured knowledge graphs that can be queried for specific information. This segmentation allows the system to retrieve only the relevant portions of information needed to answer a question, rather than processing entire text documents.
Solution Approach 2:
The patent extracts and separates the knowledge representation from the training data requirement by using knowledge graphs to store structured information. This allows the system to leverage existing structured knowledge without requiring extensive labeled training data for each specific question answering task.
4Adaptability or versatility
If unstructured text documents are used for training, then data availability is improved, but storage resources are extensively required
Solution Approach 1:
The patent changes the structural parameters of data representation by converting unstructured text into structured knowledge graphs. This transformation maintains the availability of information while significantly reducing storage requirements through efficient structured representation and indexing.
Data Source
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.