Exploring the Effectiveness of Sampling Techniques on Knowledge Graphs
Start Date
7-8-2026 10:15 AM
End Date
7-8-2026 10:30 AM
Location
ALT 205
Abstract
Knowledge graphs are becoming more popular for data storage and analytics, and their sizes are increasing rapidly due to the rate at which existing data is being transformed to the knowledge-graph format and the rate at which new knowledge is being discovered. Although larger graphs can potentially provide more insights, existing analytical tools are unable to keep up with their growing sizes. To address this efficiency problem, this project focuses on knowledge-graph sampling, which produces smaller graphs to analyze. To maintain the quality of the downstream analytical results, representative samples are needed. While many sampling techniques exist and have been studied for simple networks, their effectiveness on knowledge graphs has been largely unexplored. Knowledge graphs present unique challenges as many features of the original graphs must be maintained in representative samples, e.g., node and edge type distributions, node degrees, etc. The goal of this project is to explore the application of existing sampling techniques to knowledge graphs. Experiments have been conducted to characterize the effectiveness of each sampling method on maintaining key knowledge-graph properties. The results indicate some key benefits and drawbacks of each approach to enable more informed sampling.
Exploring the Effectiveness of Sampling Techniques on Knowledge Graphs
ALT 205
Knowledge graphs are becoming more popular for data storage and analytics, and their sizes are increasing rapidly due to the rate at which existing data is being transformed to the knowledge-graph format and the rate at which new knowledge is being discovered. Although larger graphs can potentially provide more insights, existing analytical tools are unable to keep up with their growing sizes. To address this efficiency problem, this project focuses on knowledge-graph sampling, which produces smaller graphs to analyze. To maintain the quality of the downstream analytical results, representative samples are needed. While many sampling techniques exist and have been studied for simple networks, their effectiveness on knowledge graphs has been largely unexplored. Knowledge graphs present unique challenges as many features of the original graphs must be maintained in representative samples, e.g., node and edge type distributions, node degrees, etc. The goal of this project is to explore the application of existing sampling techniques to knowledge graphs. Experiments have been conducted to characterize the effectiveness of each sampling method on maintaining key knowledge-graph properties. The results indicate some key benefits and drawbacks of each approach to enable more informed sampling.
