Knowledge Graph QA for Missing Entity Answers
Find Innovative SolutionsGenerate Solutions
Solution Overview
Problem
Traditional question answering (QA) systems fail to properly leverage entity-relation data from knowledge graphs, leading to missed information and incorrect answers, especially when the required entity is not present in the same passage as the event referenced in the question.
Innovation Solution
The approach generates a question knowledge graph and compares it to passage knowledge graphs, expanding the graph by adding neighbor entities using external data, and computes similarity scores to identify missing entities and boost candidate answer scores, incorporating new candidate answers and enhancing existing ones through a Candidate Answer Generator and Scorer phase.
Engineering Contradictions & Design Principles
Engineering Contradiction Analysis
1Reliability
If traditional QA systems search for entity names in document text, then the system can find answers when entity names are present, but the system fails to find correct answers when the required entity is not present in the same passage as the event referenced in the question
Solution Approach 1:
The patent transitions from traditional text-based entity searching to knowledge graph-based entity reasoning. By representing entities and their relationships in a structured knowledge graph format, the system can traverse relationships to find entities that are not explicitly mentioned in the passage but are connected through relational paths, thus solving the problem of missing entity information.
Solution Approach 2:
The knowledge graph serves as an intermediary between the question and the document corpus. Instead of directly searching for entity names in text, the system uses the knowledge graph to infer entity identities through relationship traversal, acting as a mediator that connects question intent with available document information even when direct entity mentions are absent.
2Adaptability or versatility
If QA systems use entity disambiguation based on entity types, then the system can handle ambiguous entity mentions, but the system fails to properly leverage entity-relation data from knowledge graphs
Solution Approach 1:
The knowledge graph structure provides a universal framework that serves multiple functions: it stores entity information, represents relationships between entities, enables disambiguation through contextual relationship analysis, and supports reasoning across different question types. This multi-functional approach replaces separate mechanisms for entity disambiguation and relationship analysis.
Solution Approach 2:
The system pre-processes document corpora into knowledge graph structures before answering questions, organizing entity-relation data in advance. This preliminary construction of knowledge graphs enables efficient querying and reasoning during the question-answering phase, allowing the system to leverage pre-computed relationship paths rather than computing them on-demand.
3Productivity
If traditional approaches search for president names in document text, then the system can find answers when names are present, but the system cannot find answers when no such president name exists in the same passage
Solution Approach 1:
The system moves from one-dimensional text keyword matching to multi-dimensional knowledge graph traversal. By representing entities and their relationships in a structured graph format, the system can explore multiple relational paths to identify entities that satisfy question constraints even when their names are not explicitly mentioned in the passage, thus maintaining both efficiency and correctness.
Solution Approach 2:
The knowledge graph-based approach provides feedback mechanisms where the system can verify entity identities through relationship consistency checks. When searching for entities like presidents, the system can use relationship paths (e.g., president -> visited -> country, president -> signed -> treaty) to verify candidate entities, providing feedback that confirms or refutes potential answers without requiring explicit name mentions.
Data Source
AI summary
An approach is provided that receives a question at a question-answering (QA) system. A number of passages are identified that are relevant to the received question. A question knowledge graph is generated that corresponds to the question and a set of passage knowledge graphs are also generated with each passage knowledge graph corresponding to one of the identified passages. Each of the passage knowledge graphs are compared to the question knowledge graph with the comparison resulting in a set of knowledge graph candidate answers (kgCAs). A set of candidate answers (CAs) is computed by the QA with at least one of the CAs being based on one of the kgCAs.