Graph Query Optimization via Pattern Frequency Prediction

The patent transforms the graph pattern matching problem into a parameter-based frequency prediction problem. Instead of using histogram-based methods that require fixed schemas, the system predicts pattern frequencies by calculating occurrence frequencies of graph patterns based on node occurrence frequencies and edge connectivity, adapting the prediction approach to the dynamic nature of graph databases.

The patent replaces the mechanical histogram construction approach with a mathematical calculation system. Rather than building histograms from data samples, the system uses occurrence frequency calculations based on node and edge properties, substituting the mechanical sampling process with mathematical predictions that work for both relational and graph databases.

The patent segments the graph pattern prediction problem into manageable components: node occurrence frequencies, edge connectivity information, and pattern composition rules. By breaking down complex graph patterns into smaller elemental units (nodes and edges) with their own frequency characteristics, the system can predict frequencies of larger patterns without storing all possible pattern instances in memory.

The patent performs preliminary calculations of node occurrence frequencies and edge connectivity metrics before actual query execution. These pre-computed base values are stored instead of complete pattern frequency tables, and final pattern frequencies are derived through calculation during query processing, reducing memory requirements while maintaining prediction accuracy.