Smart SPARQL Advisor: Guiding Users in Query Formulation with Performance Prediction

Abstract

Writing SPARQL queries is often an iterative process, where users refine queries until they meet their information needs. However, long-running query executions can lead to inefficient workflows, as users must wait idly for results — potentially without success due to strict timeouts imposed by public endpoints. In this demo, we present the Smart SPARQL Advisor (SSA), a system that integrates Query Performance Prediction (QPP) to proactively mitigate these issues. By predicting query runtimes prior to execution, SSA alerts users to potentially slow or timeout-prone queries and, when necessary, employs a large language model (LLM) guided by latent representations from the QPP model to suggest alternative query formulations. We demonstrate that SSA enables users to identify performant queries and understand performance bottlenecks, thereby reducing idle time and avoiding unproductive query executions. Through this approach, SSA fosters more responsive and resource-efficient interactions with triplestores, enhancing both user experience and triplestore utilization.