Approximated Embeddings for Out-of-Knowledge-Graph Link Prediction

Resolve Bottlenecks,
Find Innovative Solutions
Generate Solutions

Solution Overview

Problem

Traditional knowledge graph models struggle to predict new facts involving entities not present in the graph without requiring extensive retraining, which is computationally expensive and inefficient.

Innovation Solution

A method to generate approximated embeddings for out-of-knowledge-graph entities using a set of descriptory triples and an embedding matrix, allowing for plausibility predictions without modifying the knowledge graph, leveraging existing embeddings to create an on-demand predictor.

Engineering Contradictions & Design Principles

VSEngineering Contradiction Analysis

1Measurement precision

If traditional knowledge graph models are retrained to predict new facts involving out-of-knowledge-graph entities, then prediction accuracy for unknown entities is improved, but computational cost and training time increase significantly

Engineering Contradiction:
Improveprediction accuracyVSAvoidcomputational cost
Core Design Contradiction:
Measurement precisionVSUse of energy by stationary object

Solution Approach 1:

The patent creates approximated embedding vectors that copy the structure and relationships of the original knowledge graph embeddings, allowing out-of-knowledge-graph entities to be represented without retraining the entire model. This copying approach enables the model to handle new entities by generating synthetic embedding representations that mimic the patterns learned from existing entities, thus maintaining prediction accuracy while avoiding the computational burden of full retraining

Inventive Principle:
Principle #26Copying

Solution Approach 2:

The patent performs preliminary actions by pre-computing approximated embeddings for out-of-knowledge-graph entities before actual prediction tasks. The system pre-processes entity descriptions and generates embedding vectors in advance, so that when prediction is needed, the computationally intensive work has already been done, enabling fast and accurate predictions without requiring model retraining

Inventive Principle:
Principle #10Preliminary action

2Adaptability or versatility

If the knowledge graph is modified to include out-of-knowledge-graph entities, then the model can make predictions for these entities, but the complexity of the knowledge graph and retraining requirements increase

Engineering Contradiction:
Improvecapability to handle unknown entitiesVSAvoidknowledge graph structure
Core Design Contradiction:
Adaptability or versatilityVSDevice complexity

Solution Approach 1:

The patent segments the problem of handling out-of-knowledge-graph entities by separating entity representation from the core knowledge graph structure. Instead of integrating new entities directly into the knowledge graph, the system generates embedding vectors independently for out-of-knowledge-graph entities based on their descriptions, allowing the knowledge graph to remain unchanged while still enabling predictions involving new entities

Inventive Principle:
Principle #1Segmentation

Solution Approach 2:

The patent introduces approximated embedding vectors as an intermediary between out-of-knowledge-graph entities and the knowledge graph prediction model. These embedding vectors serve as a bridge, translating entity descriptions into a format compatible with the existing model without requiring modifications to the knowledge graph structure or model architecture

Inventive Principle:
Principle #24Intermediary (Mediator)

3Reliability

If full retraining is performed to improve predictions for out-of-knowledge-graph entities, then model accuracy is improved, but training time and resource consumption increase

Engineering Contradiction:
Improveprediction reliabilityVSAvoidtraining time
Core Design Contradiction:
ReliabilityVSLoss of time

Solution Approach 1:

The system copies the embedding representation strategy from known entities to unknown entities, generating approximated embeddings that replicate the successful patterns learned during original training. This copying mechanism allows the model to generalize to new entities without requiring time-consuming retraining, maintaining prediction reliability while avoiding repeated training cycles

Inventive Principle:
Principle #26Copying

Solution Approach 2:

The patent changes the parameters by using entity descriptions and approximated embedding vectors instead of requiring full model parameter updates. The system adjusts only the embedding representations for new entities based on their textual descriptions, rather than changing all model parameters through retraining, thus achieving reliable predictions with minimal time investment

Inventive Principle:
Principle #35Parameter changes

Data Source

PatentEP4174727B1Methods and systems for approximating embeddings of out-of-knowledge-graph entities for link prediction in knowledge graph
Publication Date: 2025.07.02 ACCENTURE GLOBAL SOLUTIONS LTD
  • EP4174727B1 patent drawingFigure 1
  • EP4174727B1 patent drawingFigure 2
  • EP4174727B1 patent drawingFigure 3A~3B

AI summary

The present disclosure describes methods and systems for generating an approximated embedding of an out-of-knowledge-graph entity based on a knowledge graph. The method includes: receiving a target entity, a dataset associated with the target entity, and an embeddings space of a knowledge graph comprising a set of structured data, wherein the target entity is out of the knowledge graph and the embeddings space includes a set of vectors representing the set of structured data in the embeddings space; selecting a set of elements from the knowledge graph, each element being related to the target entity according to the dataset associated with the target entity; constructing a set of descriptory triples based on the target entity and the set of elements; obtaining an embedding matrix based on the descriptory triples and the embeddings space; and generating an approximated embedding for the target entity based on the embedding matrix.