Knowledge Graph Quantity Fact Extraction via Embedding Similarity
Find Innovative SolutionsGenerate Solutions
Solution Overview
Problem
Existing methods for extracting quantity facts from textual data, such as Qsearch, suffer from low precision beyond the top-ranked facts, failing to provide reliable information for filling gaps in knowledge bases with high recall and precision.
Innovation Solution
A computer-implemented method that determines quantity facts by analyzing textual resources, using embedding spaces and similarity metrics to select relevant tuples based on context and unit similarity, thereby improving precision and recall for adding quantity facts to knowledge graphs.
Engineering Contradictions & Design Principles
Engineering Contradiction Analysis
1Measurement precision
If Qsearch is used to extract quantity facts from textual data, then recall is improved, but precision deteriorates beyond top-ranked facts
Solution Approach 1:
The patent segments the fact extraction process into multiple independent modules: query generation module, passage retrieval module, fact extraction module, and verification module. Each module handles a specific aspect of the extraction process, allowing precision optimization at each stage while maintaining comprehensive recall through the coordinated workflow of all segments.
Solution Approach 2:
The patent introduces intermediate structures such as generated queries acting as mediators between the textual data and the knowledge base, and intermediate fact representations that undergo verification before final integration. These intermediaries enable precision filtering while preserving recall by allowing systematic validation at each transformation stage.
2Productivity
If comprehensive fact extraction is performed to improve recall, then more facts are captured, but computational resources increase
Solution Approach 1:
The patent performs preliminary actions by generating multiple candidate queries before the main extraction process and pre-retrieving passages that may contain relevant facts. This preliminary preparation enables the system to efficiently filter and process only the most promising candidates during the main extraction phase, achieving high recall without proportionally increasing computational resource consumption.
Solution Approach 2:
The patent employs partial action by extracting and verifying only those facts that meet specific confidence thresholds and relevance criteria, rather than processing all possible fact candidates. This selective approach maintains high recall for meaningful facts while reducing computational overhead by excluding low-value extractions.
3Productivity
If multiple candidate facts are considered to improve recall, then more potential facts are evaluated, but processing time increases
Solution Approach 1:
The patent implements periodic verification and filtering steps in the fact extraction pipeline, where candidate facts are evaluated at multiple staged intervals rather than all at once. This periodic processing allows the system to maintain high recall by considering multiple candidates while reducing overall processing time through early elimination of low-promise candidates and batched verification operations.
Data Source
AI summary
A device and a computer-implemented method for adding a quantity fact to a knowledge base, in particular a knowledge graph. The method includes: providing the knowledge base; providing a textual resource; providing an entity from the knowledge base; providing a relation from the knowledge base; providing a set of different units; determining a quantity comprising a unit within the set of different units that is within the textual resource depending on the entity, the relation, and the set of different units; determining a quantity fact comprising the entity, the relation, the quantity and the unit; and adding the quantity fact to the knowledge base.