Knowledge Base Incorrect Triple Detection via Embedding Clustering
Find Innovative SolutionsGenerate Solutions
Solution Overview
Problem
Existing methods for detecting inconsistencies in knowledge bases are limited in defining and verifying all types of errors, leading to potential erroneous results and incomplete error correction.
Innovation Solution
An incorrect triple detection system and method using an embedding model that selects target triples, creates sentence models, embeds them into a vector space, and applies clustering and filtering to identify and filter out incorrect triples, employing techniques like Skip-gram, Glove, TransE, TransR, NTN, and HolE for accurate detection.
Engineering Contradictions & Design Principles
Engineering Contradiction Analysis
1Reliability
If rules are defined to detect inconsistencies in a knowledge base, then inconsistent knowledge base errors can be detected, but all rules for detecting other knowledge base errors cannot be directly defined and verified
Solution Approach 1:
The patent replaces the mechanical rule-based system with an embedding model that uses vector space representations and clustering algorithms. Instead of manually defining rules to detect inconsistencies, the system embeds triples into vectors and uses unsupervised learning to identify incorrect triples, thereby substituting complex rule definitions with a more flexible machine learning approach
Solution Approach 2:
The patent changes the parameter space by transforming discrete triple data into continuous vector representations. By embedding triples into a vector space and applying clustering algorithms, the system detects incorrect triples based on vector distance and cluster distribution rather than predefined rules, enabling detection of error types that cannot be captured by traditional rule-based methods
2Measurement precision
If embedding models are used to detect incorrect triples, then detection accuracy is improved, but computational complexity and processing time increase
Solution Approach 1:
The patent performs preliminary actions by pre-computing and storing embedding vectors for all triples in the knowledge base. The embedding model and clustering structures are built in advance, allowing rapid detection of incorrect triples during query time without performing heavy computations on-the-fly, thus reducing processing time while maintaining high detection precision
Data Source
AI summary
Provided is an incorrect triple detection system including a triple selector configured to select a target triple (subject, type, object) in a knowledge base, a sampler configured to create a sentence model by connecting object triples sharing entities included in the target triple, a model builder configured to embed the sentence model into a vector space to create a training entity vector and build an embedding model, and an incorrect triple detector configured to detect an incorrect triple by inputting a test triple into the embedding model.