Graph Database Query Acceleration via Machine Learning Embeddings

The patent segments the graph dataset into training data and query data portions. Machine learning models are trained on the training data portion to create pre-computed representations (embeddings) that capture the essential patterns and relationships. During query execution, these pre-computed representations are used instead of processing the entire graph dataset, thereby segmenting the computational workload and improving query performance while maintaining data completeness.

The patent applies preliminary action by pre-training machine learning models on the graph dataset before query execution. The models are trained offline to learn node embeddings, edge predictions, and graph patterns. This preliminary training creates optimized data structures and representations that can be rapidly queried without re-processing the entire dataset, thus improving query speed while maintaining comprehensive data storage.

The patent introduces machine learning models as intermediaries between the graph database and query processing. These models act as mediators that have been pre-trained to understand complex graph patterns and relationships. When a query is executed, the intermediary model rapidly retrieves pre-computed representations and patterns rather than performing exhaustive graph traversals, thereby maintaining analytical depth while significantly reducing execution time.

The patent changes the parameter representation of graph data by transforming raw graph structures into learned embeddings and feature vectors through machine learning models. This parameter transformation converts complex relational data into optimized numerical representations that can be processed much faster while preserving the semantic meaning and analytical value, thus reducing query execution time without sacrificing result accuracy.