Weighted Causal Knowledge Graph Embeddings for Link Prediction
Find Innovative SolutionsGenerate Solutions
Solution Overview
Problem
Traditional causal discovery techniques are time-consuming and rely solely on observation data, leading to incomplete causal networks and large search spaces, and existing knowledge graph embedding algorithms do not incorporate causal weights, resulting in biased model performance.
Innovation Solution
Formulate causal discovery as a knowledge graph completion problem, using weighted causal relations to predict new links, and employ FocusE to assimilate causal weights into knowledge graph embedding models, with a Markov-based data split to minimize model bias.
Engineering Contradictions & Design Principles
Engineering Contradiction Analysis
1Reliability
If traditional causal discovery techniques are used, then causal networks can be constructed from observation data, but the process is time-consuming and results in incomplete causal networks with large search spaces
Solution Approach 1:
The patent merges knowledge graph embedding algorithms with causal discovery by formulating causal discovery as a knowledge graph completion problem. This combines the strengths of both approaches: knowledge graphs provide structured semantic relationships while embedding algorithms enable efficient link prediction, reducing both time and improving completeness simultaneously
Solution Approach 2:
The patent introduces knowledge graph embeddings as an intermediary between causal networks and embedding models. By converting causal networks into knowledge graphs with embeddings, it creates a bridge that enables efficient causal link prediction while maintaining causal relationships, thus reducing computational time and improving completeness
2Measurement precision
If existing knowledge graph embedding algorithms are used, then embeddings can be generated, but causal weights are not incorporated leading to biased model performance
Solution Approach 1:
The patent applies local quality by incorporating causal weights specifically into the edge representations of the knowledge graph. Instead of treating all relationships equally, the causal weights are embedded locally at each edge, allowing the model to capture the specific causal strength of each relationship while maintaining the overall graph structure
Solution Approach 2:
The patent changes the parameters of knowledge graph embedding algorithms to incorporate causal weights. By modifying the embedding process to include causal weight as a parameter, the model can accurately represent the strength of causal relationships, eliminating bias and improving measurement precision of causal effects
3Productivity
If causal discovery is formulated as knowledge graph completion, then additional causal links can be predicted, but the search space remains large
Solution Approach 1:
The patent applies dimensionality change by transforming the causal network into a knowledge graph with embedded representations. This moves the problem from a high-dimensional search space of all possible causal links to a lower-dimensional embedding space where link prediction can be performed efficiently using vector operations, thus reducing search space complexity while maintaining prediction accuracy
Data Source
AI summary
Causal discovery is performed using knowledge graph link prediction. Information from a causal network is transformed into a causal knowledge graph according to a mapping, the causal knowledge graph including a plurality of causal links, wherein each causal link includes a cause entity, a causal relation, an effect entity, and a causal weight indicating a relative strength of causal influence of the cause entity on the effect entity. The causal knowledge graph is converted into embeddings, where the embeddings include a latent vector space representation of the causal knowledge graph. The embeddings are trained using a subset of the causal links of the causal knowledge graph. The embeddings are used for causal discovery to predict additional causal links of the causal knowledge graph.