Cluster-Based Random Walk for Large Graph Data

The patent applies preliminary action by pre-loading graph data into memory before the random walk algorithm executes. The system reads the entire graph data from the database once, stores it in memory as a data structure (e.g., adjacency list or matrix), and then performs random walk operations directly on the in-memory data. This eliminates repeated database access during the random walk process, significantly improving processing efficiency and reducing database access time.

The patent introduces memory as an intermediary between the database and the random walk algorithm. Instead of directly accessing the database during random walk operations, the system uses an in-memory data structure as a mediator to store graph data. This intermediary layer allows fast random access to nodes and edges during the algorithm execution, reducing the burden on the database and improving overall processing speed.

The patent applies segmentation by dividing the large-scale graph data into manageable chunks or blocks that can be loaded into memory. The system may partition the graph data by nodes or edges, process segments independently, and combine results. This segmentation approach enables efficient processing of large datasets that would otherwise exceed memory capacity or cause performance degradation in traditional random walk implementations.

The patent changes the parameter of data storage location from database to in-memory structure, fundamentally altering the system's performance characteristics. By changing where the graph data resides (from persistent storage to volatile memory), the system achieves O(1) or O(log n) access time for nodes and edges during random walk operations, making it scalable to large datasets without proportionally increasing database access complexity.